TL;DR: In 2017, Illia Polosukhin (@ilblackdragon) co-authored the paper that invented transformers. The architecture powers ChatGPT, Claude, and every major AI system. Months later, he left Google to build an AI company (@near_AI). That company needed to pay workers around the world. Traditional payment rails failed. Blockchain solved the problem. Seven years later, the same insight drives @NEAR_Intent. At NVIDIA GTC 2026, he argued that classical economic models need revision for agentic participants.

The same person who helped create the AI architecture that powers ChatGPT is now building the coordination layer for the AI Economy that AI agents will use to transact.

The Paper That Changed Everything

Before there was ChatGPT, before there was Claude, before there was Gemini, there was a paper.

“Attention Is All You Need,” published in June 2017 by eight Google researchers, introduced the transformer architecture. The “T” in GPT stands for transformer. Every major language model, every AI assistant, every system that seems to understand and generate human language traces back to this paper.

Illia Polosukhin was one of those eight researchers.

At NVIDIA’s GTC conference in March 2024, Jensen Huang brought seven of the eight authors on stage. “Everything that we’re enjoying today,” Huang said, “can be traced back to that moment.”

The “Transformer 8,” as they’ve become known, have all left Google. Noam Shazeer co-founded Character.AI. Aidan Gomez co-founded Cohere. Lukasz Kaiser joined OpenAI. Each took the transformer architecture in different directions.

Polosukhin took it somewhere unexpected: blockchain.

The Payment Problem

The path from transformers to blockchain wasn’t ideological. It was practical.

“We had people around the world, mostly students from China, Russia, Ukraine, etc., who were doing kind of small tasks for us,” Polosukhin explained in an interview with Sifted. “Paying these people was actually a pretty painful process. Sending money from the US to China is complicated, Russia and Ukraine as well.”

After leaving Google in 2017, Illia and Alexander Skidanov (@AlexSkidanov) started NEAR AI an AI company focused on program synthesis. They were teaching machines to write code from natural language descriptions. The technical work required data labeling at scale. Students around the world annotated examples, provided feedback, and trained the models.

The work was distributed. The payments weren’t.

Wire transfers cost $25–50 per transaction. ACH takes days and doesn’t work internationally. PayPal and traditional payment processors add friction and fees that make micropayments impractical. Sending $50 to a student in Ukraine for a day’s work shouldn’t cost $30 in fees and take a week to arrive.

They ended up looking at blockchain as a solution for their own problem.

From AI Company to Protocol

The pivot wasn’t abandoning AI. It was recognizing that AI needed infrastructure that didn’t exist yet.

@NEARProtocol launched in 2020 with a focus on usability: human-readable addresses (alice.near instead of 0x7a23…), cheap transactions, and developer-friendly tooling. Sharding for scalability, account abstraction for flexible permissions. All of it served one goal: making blockchain accessible enough that normal people could use it.

But the AI thesis never disappeared. It went underground, waiting for the models to catch up.

By 2023, they had. ChatGPT showed that transformer-based systems could handle complex tasks. Agents became plausible. The question Polosukhin had hit in 2017, how distributed AI systems coordinate and transact, suddenly mattered to everyone.

The Intents Architecture

At Solana Breakpoint 2025, Illia connected the threads:

“Before jumping into the blockchain rabbit hole, I was an AI researcher. I worked at Google Research on effectively deep learning, machine understanding. And I’m one of the co-authors of ‘Attention Is All You Need,’ which is a paper that introduced transformers, T and GPT.”

“Now in 2017, I left Google to start an AI company and we quickly realized that we actually need a blockchain so that we can coordinate and pay people around the world. That’s how we got to NEAR Protocol in 2018.

The insight from 2017 scales directly. If paying human data labelers required blockchain infrastructure, paying AI agents will require it even more.

Here’s how NEAR Intents works: users express desired outcomes. Swap this token for that one. Book this flight. Source this product. They don’t specify how. Solvers compete to fulfill the intent. Settlement happens on-chain.

The architecture has processed over $15B in volume across more than 35 chains. Trust Wallet, Ledger, Brave Wallet, Infinex, Zodl, CowSwap, and dozens of other interfaces have integrated.

Why Blockchain Was Required

The presentation at Solana Breakpoint made the point clearly: Cross-chain has been extremely painful. And if we look out, the reality is most users have been sitting on centralized exchanges kind of historically for all of the time that blockchain existed. It’s been too hard to use.

Bridges fail constantly. Gas costs confuse users. Wallets struggle to support every chain. Every blockchain became an island.

NEAR Intents solves this by abstracting execution: You can actually express the outcome you want without needing to figure out how exactly the execution’s gonna work.

The same abstraction applies to AI agents. An agent shouldn’t need to know which chain has the best liquidity for a particular swap. It shouldn’t manage gas tokens across twenty networks. It should express what it wants and let specialized solvers compete to deliver.

This is extremely important. Because solvers can source liquidity from a bunch of different sources: There’s gonna be market makers that are concentrating on Binance. There can be some prop trading firms that are taking a reverse position

→ There’s no MEV because this contract between users and the counterparty is done off-chain.

The Technical Foundation

Three pieces of infrastructure make this work.

1. Multi-party computation (MPC) handles custody of assets across different chains without centralized control. No single party holds complete keys. Threshold signatures require multiple parties to authorize transactions.
2. Chain signatures allow NEAR accounts to sign transactions on Bitcoin, Ethereum, Solana, Zcash, and +30 other chains without bridges or wrapped assets.
3. NEAR’s sharded infrastructure allows for massive throughput and private transactions — up to 1m transactions per second.

The result: Effectively providing a centralized exchange experience to self-custodial users and the rails for agentic commerce.

Expanding Beyond Trading

Trading was the first use case because the pain was most obvious. But intents extend furthe

The NEAR Intents infrastructure is not just for trading: express a complex intent in natural language, let AI agents propose solutions, settle commitments on-chain. You say where you want to go. Agents figure out how to. This then gets extended to e-commerce, to actual contractual service work. So really intents is this new paradigm to find the counterparty and get agreement with them and settle it and execute it across different avenues.

The pattern is consistent: blockchain networks work better when connected through intent-based infrastructure than when isolated behind incompatible bridges.

The AI-Native Endgame

The vision connects back to transformers.

In March 2024, Jensen Huang put Polosukhin on the GTC stage as a co-inventor of the transformer. Two years later, at GTC 2026, Polosukhin returned to present something different: a formal model for agentic markets. Not AI research this time. Infrastructure.

His argument was specific. Today’s markets were designed for humans. They’re full of irrational behavior, bias, expensive legal overhead, and rent-seeking middlemen. When AI agents transact, they optimize on outcomes. They don’t need brokers, lawyers, or intermediaries to enforce trust. They need programmatic escrow, intent-based matching and settlement, and agentic dispute resolution. Purpose-built primitives, not adaptations of existing systems. The implication: classical economic models, built on assumptions about human rationality and human friction, need revision for agentic participants.

The GTC talk also introduced IronClaw, a secure open-source agent runtime built to protect users from prompt injection and data leakage. Agents run inside Confidential Virtual Machines (CVMs), inference happens inside hardware-enforced private enclaves, and every execution comes with cryptographic attestations users can verify. Only the user and their agent can see credentials, data, and context. This is what “user-owned AI” means in practice: not a marketing phrase, but a specific technical guarantee.

The broader point: these same primitives (agents voting to allocate budget, measuring outcomes against milestones, reallocating automatically when targets are missed) extend beyond commerce to public goods funding and government spending. The persistent problem with public institutions is lack of transparency, clear attribution, and ROI measurement. Agents can fix the accountability problem at the infrastructure level.

Polosukhin helped invent the architecture that makes AI agents possible. Then he discovered that AI systems need payment infrastructure that traditional finance can’t provide. He built that infrastructure. Now AI agents are capable enough to use it.

What NEAR is trying to build is really this unified liquidity layer where users can plug in into any kind of asset beyond blockchains into Web2.

The future he describes: users interact through any wallet (Solana, NEAR, Ethereum, Bitcoin) with any application. AI agents handle coordination. Blockchain provides settlement. The execution machinery disappears.

This is what the payment problem from 2017 pointed toward. Distributed intelligence needs trustless coordination. The co-author of transformers is building it.

Conclusion

The path from “Attention Is All You Need” to NEAR Intents isn’t obvious until you trace it.

• 2017: Polosukhin co-invents transformers at Google. 2017: Leaves to build an AI company.
• 2017–2018: Discovers that paying distributed workers requires blockchain.
• 2018–2020: Builds NEAR Protocol as usable blockchain infrastructure.
• 2020–2024: Develops chain signatures, MPC custody, intent architecture.
• 2024–2026: AI agents become capable. The infrastructure is ready. Illia returns to GTC to present the market primitives.

The same person who helped create the AI architecture that powers ChatGPT is now building the coordination layer that AI agents will use to transact.

That’s not coincidence. It’s the completion of a thesis that started when an AI researcher tried to pay students in Ukraine and found that the financial system couldn’t handle it.

Blockchain solved the payment problem. Intents scale it to AI-native commerce.

The co-founder of modern AI is building its economic infrastructure.

Sources

• Illia Polosukhin, Solana Breakpoint presentation (2025) I
• llia Polosukhin, “Market Primitives for an Agentic Economy” [S82411], NVIDIA GTC (March 18, 2026)
• “Attention Is All You Need,” Vaswani et al., Google (2017): arxiv.org/abs/1706.03762
• CNBC interview with Katie Tarasov (June 2024) Sifted interview: “Why is one of the founding fathers of generative AI all in on Web3?”
• NEAR Intents analytics: dune.com/near/near-intents

Series Context — This is Part 5 of a series on NEAR Intents and the AI economy

Disclaimer: The information presented herein has been provided by third parties and is made available solely for educational and general information purposes. Nothing in this newsletter should be construed as legal, tax or investment advice. This post might not reflect all current updates to applicable laws, regulations or guidance. The authors disclaim any obligation to update this post and reserve the right to make any changes to this post without notice. In all cases, persons should conduct their own investigation and analysis of the information in this newsletter. Where this newsletter contains links to other sites and resources provided by third parties, these links are provided for information only and should not be interpreted as approval by us of those linked websites (or any information obtained from them) that the authors of this post do not control. Always DYOR.

Copyright © 2026 NEARWEEK AG, All rights reserved.

From Transformers to Intents was originally published in NEAR Protocol on Medium, where people are continuing the conversation by highlighting and responding to this story.

NEARWEEK Research. The Verifiable Stack, Settlement Layer.

The Verifiable Stack is a NEARWEEK research initiative. This series maps the infrastructure required for verifiable AI at institutional scale. It presents a framework, not a product. The goal is industry-wide adoption of verifiable AI infrastructure.

TL;DR: The GENIUS Act (July 2025) triggered stablecoin issuer proliferation. McKinsey projects $3.7T in stablecoin supply by 2030. Current DEX infrastructure cannot support institutional-grade swaps across dozens of stablecoin pairs. The Stablecoin Transport Protocol (STP) by Yonder Labs introduces uncollateralized lending to TEE-attested Shade Agent solvers for cross-chain swaps at $1M-$10M+ with 1–3bp fees and zero slippage. StableFlow (@0xStableFlow) is the live frontend, processing USDT across 9 chains at 0.01% fees. STP is the first production protocol where the verification layer is built into the settlement mechanism rather than bolted on afterward. The unsolved problem: the off-chain execution leg remains a black box, and no unified Verifiable Commerce Receipt connects solver attestation to settlement proof to agent decision provenance.

The GENIUS Act Broke the Stablecoin Duopoly

The GENIUS Act passed in July 2025. It created the first comprehensive US regulatory framework for stablecoin issuers, standardizing reserve requirements, reporting obligations, and compliance thresholds. The policy intent was stability. The market effect was fragmentation.

Within months of passage, banks, fintechs, payment companies, and technology platforms began announcing stablecoin products. USDC and USDT continue to dominate aggregate supply, but their share of the issuer landscape is declining as new entrants enter with regulatory cover that did not exist before the Act. PayPal’s PYUSD, Ripple’s RLUSD, World Liberty Financial’s USD1, and Hashnote’s USYC represent only the first wave. Every major US bank with a stablecoin strategy now has a regulatory pathway [1].

McKinsey’s “The Stable Door Opens” report projects stablecoin supply reaching $3.7T by 2030 [2]. That projection assumes dozens of issuers, each with meaningful supply, operating across multiple chains. The infrastructure question is not whether stablecoins will proliferate. The question is whether the rails that move them between chains can handle the resulting fragmentation.

A $1M swap of PYUSD for USD1 on a centralized exchange today moves the market by several basis points. On decentralized venues, the slippage is worse. Current daily stablecoin-to-stablecoin volume sits at roughly $3–5B on DEXs and $6–10B on CEXs [3]. Conservative projections place this at $100B daily by 2030. The infrastructure that handles the current load cannot scale to the projected load, particularly at the long tail of the stablecoin distribution where liquidity is thin and pair depth is shallow.

The Capital Efficiency Problem

Existing DEX protocols use automated market maker (AMM) designs where liquidity providers deposit paired assets into pools. Capital is distributed across a price range. A $10M liquidity pool on Curve or Uniswap might support a $500K swap without significant slippage, but the remaining $9.5M sits idle across price points that may never be touched. This is acceptable for high-volume pairs like USDC/USDT. It breaks for long-tail pairs where total liquidity is measured in hundreds of thousands, not millions.

Concentrated liquidity (Uniswap V3, Fluid) improves capital efficiency by letting providers target specific price ranges. But it introduces active management requirements that most liquidity providers cannot sustain, and it still fragments capital across range positions rather than deploying it to fill individual swaps.

Centralized exchanges offer better depth for major pairs but collapse on long-tail stablecoin pairs. They also introduce counterparty risk, custody risk, and jurisdictional fragmentation. An institutional desk executing a $5M PYUSD-to-RLUSD swap has no single venue with reliable depth.

The structural problem is clear. AMM designs spread capital. Concentrated liquidity fragments capital. Centralized exchanges centralize risk. None of these architectures are designed for a world with 30+ stablecoin issuers, each requiring deep cross-pair liquidity at institutional scale.

STP: Lending to Verified Solvers

The Stablecoin Transport Protocol (STP), developed by Yonder Labs within the Proximity Labs ecosystem, is built on a different architecture. STP is not an AMM. It is not an order book. It is a short-term uncollateralized lending protocol that lends stablecoins to cryptographically verified solvers for the explicit purpose of filling cross-chain swaps [3].

The capital efficiency gain is structural. An AMM pool of $10M supports a fraction of that in single-swap capacity because capital is distributed across a price curve. An STP lending pool of $10M supports a $10M swap because the entire pool is lent to a single solver for a single transaction. The pool is borrowed, the swap is executed, the pool is repaid. Utilization approaches 100% per transaction.

The solvers are not arbitrary counterparties. They are Shade Agents: cryptographically verified software instances running inside Trusted Execution Environments (TEEs) using NEAR AI’s Shade Agent framework and dStack infrastructure [4][5].

Three security primitives enforce the trust model.

TEE attestation is the first. Before a loan is issued, the solver submits a cryptographically signed remote attestation to the STP smart contract. The attestation proves the solver is running a specified, audited codebase inside a TEE. The contract verifies the attestation before releasing any funds. If the solver is running modified code, or code outside a TEE, the loan is rejected [3].

Escrow-first is the second. Solvers cannot borrow speculatively. To initiate a loan, the solver must present a signed NEAR Intents quote proving that a user has already deposited source funds into NEAR Intents. The repayment capital is secured onchain before the loan is ever issued. There is no window where the lender’s capital is exposed without a corresponding user deposit backing it [3].

Atomic settlement sequencing is the third. The TEE code enforces a strict, linear execution path. Borrowed funds route to an off-chain venue (Binance, Coinbase, Circle), execute the swap, withdraw the target stablecoin to the destination chain, finalize delivery through NEAR Intents, then reverse the process to repay the lending pool. The solver cannot deviate from this sequence. The code running inside the TEE is the enforcement mechanism [3].

A concrete example from the STP litepaper: a user wants to swap 1M USD1 on Base for 1M PYUSD on Solana. The user signs an intent and deposits USD1 into NEAR Intents. A Shade Agent solver presents the signed intent to the STP contract. STP verifies the solver’s TEE attestation, confirms the escrow deposit, and lends 999,900 USDC. The solver routes to a CEX, purchases PYUSD, withdraws to NEAR Intents on Solana, and finalizes the swap. The user receives 999,900 PYUSD. The solver converts the received USD1 to USDC and repays the pool. Lenders earn approximately $100 on the transaction [3].

Fees range from 1 to 3 basis points. Slippage is zero at the protocol level because the full pool backs each swap. The solver absorbs CEX-side slippage, which for major pairs at institutional size is typically sub-basis-point.

StableFlow: Live in Production

@0xStableFlow: (app(dot)stableflow(dot)ai), built by @DapDapMeUp, is the production frontend for the STP protocol. It is live and processing transactions. Current coverage: USDT across 9 chains (Ethereum, Arbitrum, Polygon, BNB Chain, Optimism, Avalanche, Solana, Tron). Transaction fees: 0.01%. Minimal slippage up to $1M [6].

The roadmap includes USDC, USD1, and USDH support, expanding the pair matrix as new stablecoin issuers achieve sufficient CEX depth for the solver execution leg [6].

StableFlow launched in October 2025, covered by Phemex and Coinspeaker [6]. The significance is not the frontend itself but what it proves: the STP mechanism works in production. TEE-attested solvers are borrowing, executing, and repaying in live market conditions. The architecture is not a whitepaper. It is shipping.

The Verifiable Stack Connection

STP is architecturally significant for the Verifiable Stack because the verification layer is not an add-on. It is the protocol.

TEE attestation maps directly to the Models/Computation pillar. The solver’s execution is cryptographically attested before any capital moves. The STP contract can verify that the solver is running the correct code in a genuine TEE before issuing a loan. This is verifiable computation applied to financial execution rather than LLM inference, but the primitive is identical: prove that a specific computation ran in a trusted environment.

The escrow-first requirement maps to the Data pillar. The signed NEAR Intents quote serves as cryptographic proof that user funds exist. No speculation. No oracle dependency for the core swap. For same-peg stablecoin swaps (USDC to PYUSD, both nominally $1.00), the Oracle Problem is sidestepped entirely because the price relationship is defined rather than discovered.

Atomic settlement sequencing maps to the Governance pillar. The TEE enforces a strict execution path that functions as an embedded compliance mechanism. The solver cannot deviate. The code is the audit trail. This is policy-as-code applied to financial settlement, one of the Verifiable Stack’s governance primitives implemented at the protocol level.

The Black Box Leg

STP solves verification at the protocol layer. It does not solve verification at the execution layer.

Execution quality attestation is the first gap. The TEE proves the solver ran the correct code. It does not prove the solver achieved best execution at the off-chain venue. The CEX leg (Binance, Coinbase, Circle) is opaque. The swap executed, but was the fill price optimal? Was the order routed efficiently? Was there front-running? No attestation exists for what happens inside the centralized exchange [3].

The FX Oracle Problem is the second. Stablecoin-to-stablecoin swaps at approximately 1:1 parity avoid oracle dependency. But STP’s roadmap includes tokenized forex (JPY/USD, EUR/USD) and eventually tokenized RWAs (treasuries, bonds, stocks). At that point, the Data pillar’s Oracle Problem returns in full. The exchange rate was signed by someone. Cryptography can prove the signature is valid. It cannot prove the rate was fair at the moment of signing [3].

Agent decision provenance is the third. When AI agents, not human users, submit intents to STP, the question shifts from “did the swap execute correctly” to “should the swap have been initiated at all.” Which model decided to swap this stablecoin for that stablecoin at this moment? What data informed the decision? Was the model compromised by prompt injection? NEAR AI’s MPC-of-TEEs can verify the inference. STP’s Shade Agents can verify the execution. But the attestation chain connecting verified inference to STP solver attestation to NEAR Intents settlement does not exist as a unified standard [7].

The Unified Verifiable Commerce Receipt remains the missing artifact. STP proves solver execution. NEAR Intents proves settlement. NEAR AI proves inference. No protocol connects these three attestations into a single verifiable artifact that a regulator, auditor, or institutional counterparty can inspect end-to-end.

Regulatory Convergence

The regulatory pressure on stablecoin settlement is compounding from multiple directions simultaneously.

The GENIUS Act created the framework that enabled stablecoin fragmentation but does not mandate execution attestation. It standardizes what issuers must hold in reserve and how they must report. It says nothing about how swaps between stablecoins are verified at the transport layer [1].

The Bank of England’s systemic stablecoin consultation (November 2025, closed February 2026) proposed holding limits of 20,000 GBP for individuals and 10,000,000 GBP for businesses, with wholesale financial market settlement exempt. The BoE’s framework contemplates a world where stablecoins are systemically important payment instruments [8].

MiCA (Markets in Crypto-Assets) continues to impose compliance requirements on stablecoin issuers operating in the EU. The regulation addresses issuance, reserve management, and consumer protection but does not address cross-chain transport or execution attestation [8].

The EU AI Act intersection is the critical convergence point. When AI agents route stablecoin flows on behalf of users, these transactions become candidates for high-risk AI system classification under Article 6. Article 12 mandates automatic logging sufficient to reconstruct decisions. Article 13 mandates transparency provisions. Enforcement begins August 2, 2026 [9].

The implication is direct. Stablecoin flows executed by AI agents will require both financial compliance (GENIUS Act, MiCA, BoE framework) and AI compliance (EU AI Act). STP’s TEE attestation satisfies a portion of both: it proves what code ran and that the execution followed a specified path. The Verifiable Commerce Receipt that connects inference attestation to execution attestation to settlement proof is what satisfies both regimes completely. Neither STP nor any competitor provides this end-to-end today.

Who Is Building This

STP / Yonder Labs (Proximity ): TEE-attested uncollateralized lending for stablecoin transport. Three security primitives. $1M-$10M+ capacity per swap. Live via StableFlow.

StableFlow / DapDap (@0xStableFlow): Production frontend. USDT across 9 chains. 0.01% fees. Roadmap: USDC, USD1, USDH.

@NEAR_Intents: ($13B+ all-time volume, 18.2M swaps, 130+ assets, 25+ chains): Settlement and escrow layer that STP depends on for the escrow-first security primitive [10].

NEAR AI (@near_ai): MPC across distributed TEEs. Shade Agent framework that STP solvers run on. D-Stack TEE infrastructure [4][5].

Chainlink CCIP: Cross-chain messaging with Risk Management Network and programmable token transfers. Does not address stablecoin-specific capital efficiency.

Eco Protocol: Stablecoin-specific, intent-based transport. Similar problem framing. Different execution architecture (no TEE attestation).

Circle CCTP: Native USDC burn-and-mint transfers. Chain-specific, single-issuer. Does not address cross-issuer stablecoin swaps.

Bridge / Stripe ($1.1B acquisition): Fiat-stablecoin rails. Enterprise-focused. Addresses on-ramp/off-ramp, not stablecoin-to-stablecoin transport.

USDT0 / LayerZero: Burn-and-mint OFT standard across 18+ chains. Single-issuer transport (Tether). Does not address cross-issuer pairs.

The gap everyone shares: no protocol connects settlement proof, execution attestation, and agent decision provenance into a single verifiable artifact. STP comes closest by embedding TEE attestation into the settlement mechanism. The CEX execution leg and agent reasoning layer remain unattested.

Connection to the Stack

Stablecoin settlement is where the Verifiable Stack thesis meets production economics. STP demonstrates that the Hardware pillar (TEEs), the Network pillar (attestation transport between solver and smart contract), the Data pillar (escrow-first as data integrity), and the Governance pillar (atomic sequencing as embedded compliance) can converge in a single protocol.

The remaining gap is the Application and Models pillars: when an autonomous agent decides to initiate a stablecoin swap, the reasoning behind that decision must be verifiable. STP proves the swap executed correctly. It cannot prove the swap should have happened at all.

The settlement layer now has a flight recorder for execution. It does not yet have one for decisions.

Sources

[1] GENIUS Act (July 2025). US stablecoin regulatory framework. Standardized issuer requirements, reserve obligations, reporting thresholds. Source: World Economic Forum coverage.

[2] McKinsey, “The Stable Door Opens.” Projects $3.7T stablecoin supply by 2030.

[3] Stablecoin Transport Protocol (STP) Litepaper. Yonder Labs, Proximity Labs ecosystem. Mechanism design: uncollateralized lending to TEE-attested Shade Agent solvers. Three security primitives: TEE attestation, escrow-first, atomic settlement sequencing.

[4] NEAR Shade Agents Documentation. Framework for cryptographically verified software instances running inside TEEs.

[5] dStack TEE Infrastructure. TEE infrastructure layer used by Shade Agents.

[6] StableFlow . Built by DapDap. USDT across 9 chains. 0.01% fees. @0xStableFlow

[7] NEARWEEK, “The Verifiable Stack” Series Opener. NEAR AI MPC-of-TEE architecture for verifiable inference.

[8] Bank of England, Systemic Stablecoin Consultation (November 2025, closed February 2026). Proposed holding limits: 20,000 GBP (individuals), 10,000,000 GBP (businesses). MiCA: EU regulatory framework for crypto-asset issuers.

[9] EU AI Act, Regulation 2024/1689. Article 6: Classification rules for high-risk AI systems. Articles 12–13: Record-keeping and transparency. Entry into force: August 2, 2026.

[10] NEAR Intents Dune Dashboard.

Disclaimer

This article is for educational and informational purposes only. It does not constitute legal, tax, investment, or financial advice. Token distribution mechanisms involve complex regulatory considerations that vary by jurisdiction. Protocols should consult legal counsel before implementing any tokenomics model discussed herein.

Stablecoins as the Settlement Layer was originally published in NEAR Protocol on Medium, where people are continuing the conversation by highlighting and responding to this story.

The 1990s were a transformative era in technological history, defined by the rise of the World Wide Web, the maturation of cryptographic protocols, and foundational advancements in artificial intelligence (AI). This decade saw the internet evolve from an academic curiosity to a global communication network, cryptography secure the digital frontier, and AI set the stage for its eventual renaissance. These developments laid the groundwork for the interconnected, decentralized, and intelligent systems that shape our lives today.

The Rise of the World Wide Web

The 1990s began with the introduction of the World Wide Web (WWW) by Tim Berners-Lee in 1991, a development that forever altered how people accessed and shared information. Unlike the earlier text-based internet, the WWW introduced a system of linked hypertext documents, accessible through user-friendly browsers. This innovation democratized the internet, allowing people to navigate with ease, regardless of technical expertise. Early browsers like Mosaic (1993) and Netscape Navigator (1994) opened the floodgates for a new wave of users and innovators, transforming the web into a space for commerce, collaboration, and creativity.

The WWW’s ability to hyperlink documents revolutionized how information was organized and retrieved. For the first time, individuals could seamlessly jump between related content, creating a dynamic and interactive experience. As a result, businesses, governments, and educational institutions flocked to establish a presence on the web, accelerating its adoption. Platforms like Amazon (1994) and eBay (1995) capitalized on this connectivity to pioneer e-commerce, while search engines such as Yahoo! (1994) and Google (1998) emerged to organize the rapidly growing volume of web content. This digital transformation marked the internet as not just a tool for information but a foundation for global commerce and culture.

Personal Computing and the Internet Economy

The 1990s were also defined by the widespread adoption of personal computers and their integration into the internet. Affordable and user-friendly PCs brought millions online, transforming the internet from a niche tool to a necessity. The development of the Transmission Control Protocol/Internet Protocol (TCP/IP) as the universal standard enabled seamless communication between disparate networks, creating the interconnected internet we know today.

E-commerce platforms like Amazon and eBay flourished, leveraging cryptographic protocols to secure online transactions. Meanwhile, the rise of email, facilitated by innovations like SMTP (Simple Mail Transfer Protocol), revolutionized communication, making instant global correspondence a reality. This shift to a digital-first economy highlighted the importance of secure, scalable infrastructure — a demand met by the cryptographic and networking breakthroughs of the era.

Cryptography: Securing the Digital Frontier

As the internet expanded, the need for secure communication and data protection grew urgent, and cryptography stepped up to meet the challenge. The introduction of Secure Sockets Layer (SSL) by Netscape in 1995, later replaced by Transport Layer Security (TLS), enabled encrypted communication between web browsers and servers. This technology secured e-commerce transactions, online banking, and personal data exchanges, laying the foundation of trust for the burgeoning internet economy.

TLS operates by encrypting data exchanged between two endpoints — such as a web browser and a server — ensuring confidentiality, integrity, and authentication. When a connection is initiated, the TLS handshake establishes a secure session by verifying identities using digital certificates and agreeing on encryption algorithms. Public-key cryptography is employed during the handshake to securely exchange a session key, which is then used for faster symmetric encryption of the actual data transfer.

TLS ensures data integrity using Message Authentication Codes (MACs), which detect tampering or corruption. Authentication typically relies on trusted Certificate Authorities (CAs) that issue certificates verifying the server’s identity. A key strength of TLS is its layered design, which operates independently of application protocols like HTTP, making it versatile for securing various types of communication.

Despite its robust security model, TLS has weaknesses. It is reliant on the trustworthiness of Certificate Authorities — if a CA is compromised, attackers can issue fraudulent certificates, enabling man-in-the-middle attacks. Misconfigurations, outdated versions of TLS, or vulnerabilities in its implementation (e.g., Heartbleed in OpenSSL) can also expose connections to risks. Moreover, while TLS protects data in transit, it does not inherently secure data at rest or address vulnerabilities in the endpoints themselves. These limitations highlight the need for comprehensive security measures alongside TLS to safeguard modern communication systems.

The SHA-1 Hash Protocol

SHA-1 is a cryptographic hash function designed to convert any input into a fixed 160-bit output, commonly used for verifying data integrity. Here’s a high-level overview of how it works:

1. Message Preprocessing: The input message is padded with extra bits and its total length is appended to ensure the message fits a specific size requirement. This prepares it for processing in 512-bit chunks.
2. Message Parsing: The padded message is divided into blocks, each of which is processed independently through the algorithm’s steps.
3. Initialization: SHA-1 uses five 32-bit variables, initialized with predefined constants, to store intermediate and final hash values.
4. Processing Blocks: Each 512-bit block is expanded into 80 smaller chunks, which are used in an iterative process. During this process, the hash values are updated using logical operations, bit shifts, and predefined constants.
5. Updating the Hash: After processing each block, the algorithm updates the intermediate hash values by combining the results from the previous steps.
6. Final Output: Once all blocks are processed, the final hash is produced by concatenating the five variables. The result is a unique 160-bit hash for the given input.

Weaknesses of SHA-1

SHA-1 is now considered insecure due to vulnerabilities that allow attackers to find collisions (two inputs that produce the same hash). This has led to its deprecation in favor of stronger algorithms like SHA-256. Despite its weaknesses, SHA-1 played a foundational role in the development of cryptographic hash functions.

AI Foundations: A Decade of Quiet Progress

While AI faced lingering skepticism from the “AI Winter” of the 1970s and 1980s, the 1990s saw steady progress in key areas. The rise of the internet provided unprecedented access to large datasets, an essential ingredient for training machine learning models. Algorithms like support vector machines (SVMs) and ensemble methods emerged, improving the ability to classify and predict patterns in data.

Advances in speech recognition and image processing hinted at the transformative potential of AI, even as researchers grappled with limited computational power. A major milestone came in 1997 when IBM’s Deep Blue defeated chess champion Garry Kasparov, proving the efficacy of specialized AI systems. This victory renewed public interest in AI and demonstrated the value of domain-specific problem-solving.

The Open-Source Movement: Collaboration as Innovation

The 1990s solidified open-source software as a powerful driver of technological innovation, fostering a collaborative ethos that continues to influence the development of critical systems today. Among the most significant milestones was the release of the Linux kernel by Linus Torvalds in 1991. This lightweight yet versatile operating system, freely available for modification and redistribution, quickly became the backbone of server infrastructure worldwide. Developers across the globe contributed to its growth, refining Linux into a robust, scalable system that powered everything from enterprise applications to emerging web technologies.

Linux’s success went hand-in-hand with the rise of tools and platforms that supported open collaboration. In 1995, the Apache HTTP Server was launched, becoming the dominant web server software of the era. Apache was pivotal in enabling the explosive growth of the World Wide Web, serving as the foundation for countless websites and online services. Its open-source nature encouraged developers to customize and improve the software, proving that collaborative development could yield technologies reliable enough for mission-critical applications.

Open-source development in the 1990s also extended to programming languages and software tools. Languages like Python (first released in 1991) gained traction due to their simplicity and versatility, enabling a wide range of applications from web development to scientific computing. Similarly, the version control system CVS (Concurrent Versions System) became a cornerstone for managing collaborative software projects, allowing developers to contribute seamlessly to large-scale efforts.

These developments exemplified the transformative potential of open-source software, providing a model of transparency, accessibility, and collective problem-solving. The ethos of shared innovation established in the 1990s not only accelerated technological progress but also laid the groundwork for decentralized systems like blockchain and collaborative frameworks for AI research, proving that open collaboration could achieve what no single entity could accomplish alone.

The Legacy of the 1990s

The innovations of the 1990s were nothing short of revolutionary, shaping the digital landscape of the 21st century. The World Wide Web transformed how information was accessed and shared, cryptography secured the foundations of digital communication, and AI research laid the groundwork for the breakthroughs of the 2000s. The rise of open-source software exemplified the power of collaboration, enabling the development of technologies that continue to define our interconnected world.

As we reflect on this pivotal decade, it is clear that the advances of the 1990s were not just steps forward — they were leaps that propelled us into the digital age, setting the stage for the cryptographic, AI, and open-source revolutions that followed.

About NEARWEEK

NEARWEEK is the ultimate destination for all things related to NEAR. As the official NEAR Protocol newsletter and community platform, NEARWEEK is the one-stop media for everything happening in the NEAR ecosystem.

NEAR Newsletter | Twitter

About NEAR Protocol

NEAR is on a mission to onboard a billion users to the limitless possibilities of Web3 with chain abstraction. Leveraging its high-performance, carbon-neutral protocol, which is swift, secure, and scalable, NEAR offers a common layer for browsing and discovering the Open Web.

NEAR AI | What is Chain Abstraction? | Twitter

The 1990s: The Internet Revolution, Cryptographic Breakthroughs, and AI Foundations was originally published in NEAR Protocol on Medium, where people are continuing the conversation by highlighting and responding to this story.

TL;DR:
@near_intents is processing billions in cross-chain volume with parabolic growth. If its trajectory holds, it is well on track to become a serious contender as the default liquidity layer for an agentic, multi-chain economy.

Navigating the multi-chain world of web3 is a mess. Swapping tokens across blockchains involves navigating bridges, wrapped assets, and conflicting networks, a complex and often costly process. What if you could simply state what you want and let someone else figure out the best way to make it happen in a completely trustless manner?

Enter NEAR Intents. What began as an experimental routing layer is rapidly maturing into a unified backbone for trustless value exchange. By abstracting away the technical complexity of multi-chain execution, it’s positioning itself as the universal protocol for transactions, whether initiated by a human or an AI agent.

By the Numbers: Parabolic Growth

The data from the NEAR Intents Dune Analytics board tells a compelling story of accelerating adoption. In less than a year, NEAR Intents has processed over $1.8 billion in volume through 3.6 million swaps, supporting 121 different assets.

The most striking insight isn’t the total, it’s the velocity.

• 30-day volume: $796.9M (44% of all-time volume)
• 7-day volume: $384.5M (Half of the 30-day volume)
• 24-hour volume: $42.9M
• Unique Users: 17,092 (24h) | 89,934 (7d) | 237,024 (30d)

This surge isn’t driven by a single chain or token. Rising activity on Ethereum, Base, TRON, Arbitrum, and NEAR itself confirms the protocol’s growth is ecosystem-wide. As noted by @resdegen:

What Are NEAR Intents?

In traditional blockchain transactions, you must specify every step of the process, just like meticulously following a recipe. NEAR Intents flips this model. You simply declare a desired outcome: “swap 100 USDC on Arbitrum for native Bitcoin” and a network of solvers competes to find the best path to fulfill your request.

The Engine Room: How It Works

The architecture is built for a multi-chain reality, resting on three pillars:

1. Users & Applications: Wallets and dApps where users formulate and broadcast their intents.
2. Solvers: A competitive, off-chain network of market makers who source liquidity and bid to fulfill requests.
3. The Verifier Contract: A smart contract on NEAR that acts as a trustless escrow. It verifies and settles transactions atomically, meaning everything happens as agreed, or nothing happens at all.

This is powered by two key technologies from NEAR Protocol:

• Chain Signatures: This breakthrough allows a NEAR account to securely sign transactions on any supported blockchain (Bitcoin, Ethereum, Solana, etc.) without relying on risky bridges. You maintain full, non-custodial control of your assets across all chains from a single account.
• Sharding: NEAR’s dynamic sharding provides the scalability to handle massive transaction volumes, a critical feature for supporting a future filled with AI-agent-driven commerce.

The result? Cross-chain settlements that finalize in 1–2 seconds, not minutes or hours.

Why It Matters: The Big Picture

This is more than a better swap engine. NEAR Intents is building critical infrastructure for the future of the web.

• For Users: It delivers a seamless, one-click experience for complex cross-chain actions, finally making multi-chain DeFi user-friendly.
• For Developers: It provides a powerful intents layer to build upon, enabling new applications that were previously too complex to engineer.
• For the AI Economy: This is the core of its potential. As autonomous AI agents need to transact (buying data, paying for compute, settling trades) they require a frictionless, trust-minimized protocol. NEAR Intents is positioning itself to be the transactional backbone for this agent-to-agent and agent-to-human economy.

The bullish outlook is supported by the data. If momentum continues, the protocol could process an additional $4–5 billion in volume in the next 90 days, pushing cumulative totals past $6 billion.

The Verdict: The Connective Tissue

NEAR Intents is transitioning from a promising experiment to essential infrastructure. It solves a critical user experience problem today while laying the groundwork for the autonomous economy of tomorrow. More and more frontier teams are already integrating or building upon the technology including

@infinex @zcash @kybernetwork @thorswap and more @chronear:

By combining a powerful intent-based architecture with NEAR’s core technical advantages, NEAR Intents isn’t just simplifying DeFi, it’s building the connective tissue for the AI driven multi-chain .

Resources

• NEAR Intents Docs
• Github
• Devhub Video presentation
• NEAR Intents Analytics Dashboard
• Reference implementation:
• Near Intents Swap

Written by: @quadron3stat3

About NEARWEEK

NEARWEEK is the ultimate destination for all things related to NEAR. As the official NEAR Protocol newsletter and community platform, NEARWEEK is the one-stop media for everything happening in the NEAR ecosystem.

NEAR Newsletter | Twitter

About NEAR Protocol

NEAR is on a mission to onboard a billion users to the limitless possibilities of Web3 with chain abstraction. Leveraging its high-performance, carbon-neutral protocol, which is swift, secure, and scalable, NEAR offers a common layer for browsing and discovering the Open Web.

NEAR AI | What is Chain Abstraction? | Twitter

Disclaimer: The information presented herein has been provided by third parties and is made available solely for educational and general information purposes. Nothing in this article should be construed as legal, tax or investment advice. This post might not reflect all current updates to applicable laws, regulations or guidance. The authors disclaim any obligation to update this post and reserve the right to make any changes to this post without notice. In all cases, persons should conduct their own investigation and analysis of the information in this newsletter. Where this article contains links to other sites and resources provided by third parties, these links are provided for information only and should not be interpreted as approval by us of those linked websites (or any information obtained from them) that the authors of this post do not control. Always DYOR.

Copyright © 2025 NEARWEEK AG, All rights reserved

The Intent-Centric Future: How NEAR Is Building the Universal Transaction Layer was originally published in NEAR Protocol on Medium, where people are continuing the conversation by highlighting and responding to this story.

Every ChatGPT reply. Every AI-generated image. Every smart contract that queries a language model. Behind it all is one invisible engine: inference.

It’s not flashy. It’s not what grabs headlines. But inference — the process of running AI models in real time — is quickly becoming the defining cost, constraint, and control point in AI’s global stack.

While the world focused on training, inference quietly became the make-or-break layer for builders. And as those costs mount, a new kind of AI infrastructure is emerging — one that looks a lot more like crypto than cloud.

Inference: The Silent Bottleneck Scaling AI

Over the past five years, the spotlight has been on bigger models and training runs: GPT-3.5, GPT-4, Claude, Gemini. But beneath the surface, inference has become the true operational bottleneck.

It’s what happens every time:

• A DAO audits its treasury with an LLM
• A wallet queries an agent to evaluate smart contract risk
• A user chats with a decentralized assistant

And unlike training — a one-time cost — inference is continuous. It’s the cloud compute tax paid with every request.

Centralized Inference: Scalable, but Costly and Opaque

Today’s inference is run on hyperscaler infrastructure — AWS, GCP, Azure — powered by expensive GPUs (A100, H100) and proprietary APIs.

Advantages:

• High throughput performance
• Enterprise-ready compliance
• Developer-friendly APIs

Drawbacks:

• Rising costs as demand and token usage grow
• Scarcity of access to top-tier GPUs
• Opaque systems with no auditability
• Platform lock-in and censorship risk

Centralized vs. Decentralized Inference: A Cost Comparison

Decentralized Inference Approaches: A Comparative Overview

Open vs. Closed Models: Control vs. Convenience

Open-source models provide:

• Full transparency and auditability
• Customization and local deployment
• Independence from proprietary APIs

Closed-source models offer:

• Higher performance and scale
• Standardized behavior
• Managed infrastructure and upgrades

Builders in crypto will recognize the parallel:
Open = sovereignty. Closed = dependency.

Verifiability and Confidentiality: AI’s Trust Layer

Inference is increasingly tied to critical decisions — in healthcare, finance, governance. And that raises a pressing question: Can we verify and protect what’s happening under the hood?

Verifiability:

• Proof-of-inference protocols
• Blockchain-based audit trails
• Zero-knowledge verification techniques

Confidentiality:

• Trusted Execution Environments (TEEs)
• Homomorphic encryption
• Confidential computing techniques

The Road Ahead: Hybrid, Tokenized, Trust-Minimized

The future of inference won’t be monolithic. It will be:

• Hybrid: Cloud + edge + decentralized nodes
• Tokenized: Incentive-aligned compute coordination
• Private-by-default: Verifiability and confidentiality embedded
• Modular: API-driven plug-and-play agent frameworks

As the cost of intelligence drops and sovereignty becomes more valuable, decentralized inference will reshape the stack — and the economics of AI itself.

Final Thought: Inference Is the New Infrastructure Layer

Inference is where AI meets compute — and where AI meets crypto.

It’s the convergence of cost, access, privacy, and trust. And for the builders working on open systems, this is the real infrastructure layer to watch.

Full Source List

• arXiv:2502.00722 — Demystifying Cost-Efficiency in LLM Serving
• arXiv:2404.14527 — Mélange: Cost Efficient Large Language Model Serving
• arXiv:2110.05014 — Cost of Decentralized Inference Under Privacy Constraints
• arXiv:2407.19401 — A Framework for Decentralized Inference
• arXiv:2306.01603 — Decentralized Federated Learning: A Survey and Perspective
• ScienceDirect — Federated Learning Cost Analysis
• Red Hat — What Is AI Inference?
• NVIDIA — Balancing Cost, Latency, and Performance in AI Inference

About NEARWEEK

NEARWEEK is the ultimate destination for all things related to NEAR. As the official NEAR Protocol newsletter and community platform, NEARWEEK is the one-stop media for everything happening in the NEAR ecosystem.

NEAR Newsletter | Twitter

About NEAR Protocol

NEAR is on a mission to onboard a billion users to the limitless possibilities of Web3 with chain abstraction. Leveraging its high-performance, carbon-neutral protocol, which is swift, secure, and scalable, NEAR offers a common layer for browsing and discovering the Open Web.

NEAR AI | What is Chain Abstraction? | Twitter

BUILDING BLOCKS: INFERENCE was originally published in NEAR Protocol on Medium, where people are continuing the conversation by highlighting and responding to this story.

Decentralized finance (DeFi) today faces a pressing challenge: billions in crypto assets remain idle on their native chains, unable to easily participate in productive financial activities due to liquidity fragmentation and complicated bridging mechanisms. Traditional methods force users to wrap tokens, moving them off their native chains and introducing friction, high fees, slow processing, and increased security vulnerabilities.

NEAR Protocol introduced Chain Signatures in 2024 to address these issues head-on, unlocking fast and secure cross-chain liquidity. Now MoreMarkets is using Chain Signatures to introduce a revolutionary model of verticalized DeFi. Unlike traditional bridging, which requires users to transfer and wrap tokens, MoreMarkets retains token security and native integrity. Users deposit tokens, such as XRP, directly into dedicated vaults on their native chains, triggering cryptographic proof creation through NEAR’s Chain Signatures. Smart contracts on destination blockchains then trustlessly validate these proofs, creating a digital representation: “More” tokens that mirror the locked assets. This model encourages broader DeFi participation without traditional bridging complexity.

Understanding NEAR’s Chain Signatures

Chain Signatures represent NEAR Protocol’s groundbreaking solution for secure cross-chain transactions. Chain Signatures enable NEAR accounts, including smart contracts, to authorize transactions on external blockchains securely. This system leverages a decentralized multi-party computation network, which combines NEAR staking with Eigenlayer ETH restakers to safeguard cross-chain interactions.

The technical infrastructure of NEAR’s Chain Signatures encompasses three core elements:

• Native Asset Vaults: Assets remain securely on their native blockchain, protected by smart contracts which trigger cryptographic signature events when deposits occur.
• Threshold Multi-Party Computation: Multiple independent “guardian nodes” manage distributed cryptographic proofs. When an asset deposit is made, guardians collaborate to generate a cryptographic proof, attesting to the locked asset’s status. The result is a system that eliminates single points of failure, isolates risks, and minimizes attack vectors inherent in traditional bridging solutions.
• Cross-chain Messaging via NFFL: Cryptographic proofs are communicated through the NEAR-enabled NFFL protocol, a secure cross-chain messaging framework, verifying asset status to receiving blockchains.

Unlike traditional bridging that requires tokens to physically move off their native chains — often causing complexity, security vulnerabilities, and fragmented liquidity — Chain Signatures keep tokens securely locked on their original blockchain. Through cryptographic proofs, Chain Signatures verify these locked states trustlessly across other blockchains, enabling asset utilization without physical token movement.

For users and developers, this means:

• Reduced Risk: By avoiding traditional multisig structures, Chain Signatures lower vulnerabilities traditionally associated with bridge exploits.
• Enhanced User Experience: NEAR’s efficient consensus and verification algorithms offer substantially reduced latency, allowing quicker, smoother cross-chain interactions.
• Increased Asset Utilization: By transforming previously idle tokens into productive assets, token holders benefit from significantly higher yields and capital efficiency.

Real-World Use Case: XRP on MoreMarkets

Consider XRP, a widely-held asset with historically low utilization rate in DeFi (0.07% AUR). With MoreMarkets and NEAR Chain Signatures, users can deposit XRP securely into a native XRP Ledger vault. This deposit triggers NEAR’s MPC-based chain signature system to cryptographically confirm XRP’s secure state. MoreXRP tokens are then minted on Solana or any other demand chain, representing the XRP locked in its home chain. Strategists deploy these MoreXRP tokens into yield-generating strategies across DeFi, funneling yield directly back to the user’s original vault.

A New Era of Global Liquidity

MoreMarkets and NEAR Protocol’s Chain Signatures embody a shift toward a more interconnected, efficient, and secure DeFi ecosystem. The integration of Chain Signatures promotes deeper liquidity pools, improved yields, and a simpler user experience. By ensuring assets remain secure on their native chains, NEAR Protocol drives a robust foundation for truly global liquidity.

Disclaimer:

This post is intended solely for informational and educational purposes. It does not constitute financial, investment, or legal advice. NEAR Protocol’s Chain Signatures and their integration with MoreMarkets and cross-chain liquidity solutions involve innovative technologies that may carry inherent risks, including but not limited to technical vulnerabilities, market volatility, regulatory uncertainty, and potential loss of digital assets. Users should independently evaluate the technology and associated risks before interacting with the protocol or related products. Past performance or technical capabilities do not guarantee future outcomes or security. The publisher makes no representations or warranties regarding the accuracy, completeness, or reliability of the information provided herein. Users are solely responsible for their own decisions and actions.

How NEAR Protocol’s Chain Signatures Power MoreMarkets and Cross-Chain Liquidity was originally published in NEAR Protocol on Medium, where people are continuing the conversation by highlighting and responding to this story.

In advance of my presentation at NVIDIA GTC today, we just published a pre-print of “Decentralized Confidential Machine Learning.” We present a novel approach to training, fine-tuning, and utilizing models and agents that allows for the decentralized, transparent properties of open source while creating a business model for developers and researchers.

Currently, using advanced AI models requires giving up privacy of users’ data and limits how models can be used. Meanwhile, open source models do not yet have a sustainable business model (not to mention, most open source AI is really just open weights). We present a decentralized system that enables the creation and deployment of LLMs and AI agents that are both open-source and monetizable; are private and verifiable while open for all to use; and preserve users’ ownership of their data and assets while enabling customized experiences that improve their well-being.

The paper presents a design for a confidential, decentralized AI cloud that combines Trusted Execution Environments or TEEs (CPU and GPU) and payment channels with a mechanism called Proof of Response to deliver service-level agreements in a decentralized environment. The DCML system is a big step towards making User-Owned AI a reality — and it makes use of new technology that has been available for less than a year.

The team went down this research path in particular because we believe TEEs are the best way to guarantee both confidentiality and verifiability together, i.e. not just a guarantee that the computation being run is what you expect, but that only this specific computation was done on this data and no other. Other potential solutions, such as ZKML, don’t offer confidentiality. And despite recent improvements, homomorphic encryption still comes with massive overhead, which presents a major challenge at scale. TEEs, on the other hand, give strong guarantees on both confidentiality and verifiability with less overhead: in fact, the bigger the model, the less overhead. They also offer the additional guarantee that only this specific computation was done on this specific data, which you don’t get from other approaches. (Side channel attacks are possible, but this is an addressable problem.).

We’re using TEE to enable verifiable and confidential VM (CVM), in collaboration with Phala Network, to enable abstraction that anyone can use. This approach unlocks some powerful benefits:

• Verifiable inference: load a model in and know your response isn’t known by whoever is running hardware or software. You also get confirmation that the model you expected to be run, was indeed run.
• Monetization of models: You can publish encrypted weights (which can only be decrypted inside the secure enclave) and then monetize the model since the secure enclave can enforce payment (streaming payments). Anytime someone wants to run inference, they attach proof of payment and get proof of compute used. Also supports monetization of user interactions without developers needing to deal with private data.
• Fine tune on top of these models with your own private data and run them on someone else’s hardware, and you can either use your model privately or monetize it for others to run.
• Decentralized training: allows a set of participants to pool funds together, train the model, and share the profits in a decentralized manner (without trust requirements).
• Enables creators to add safety restrictions to models.

Decentralized confidential machine learning introduces end-to-end confidentiality and paths to monetization for both closed source and open source AI builders while offering solutions to some of the biggest problems facing the AI ecosystem today. We believe this decentralized system opens up paths to more distributed, energy-efficient AI systems; ensures global access for participants; and makes possible new use cases for which strict data confidentiality is required, whether corporate or personal.

For more details, read the first version of the paper here, which also includes results comparing how popular models perform in TEEs versus bare metal. We welcome comments and feedback.

Decentralized Confidential Machine Learning: A Business Model for User-Owned AI was originally published in NEAR Protocol on Medium, where people are continuing the conversation by highlighting and responding to this story.

At ETHDenver 2025, the NEAR AI team introduced an exciting collaboration of forward-thinking teams including NEAR AI and Coinbase Onramp & AgentKit: the Open Agents Alliance (OAA), an initiative to deliver powerful, open source AI services to all users by combining infrastructure. With secure infrastructure powered by TEEs, an innovative and inclusive payment rail, and cutting-edge AI technology, the OAA teams’ shared mission is to ensure secure, open source, economical, and fair AI access for humanity, prioritizing user privacy and economic inclusivity for over 5.5 billion web users worldwide.

“As the global web continues to shift toward mobile users, we believe AI must be accessible to everyone,” said Illia Polosukhin, Co-Founder of NEAR AI. “In partnership with leaders in hosting, privacy, and payments, we’re building a globally distributed network that can provide AI services securely, confidentially, and at no cost to end users. This effort spans far beyond the current 350 million blockchain users, extending the power of crypto to deliver accessible AI tools to all.”

Through the OAA, participating organizations will offer end-to-end solutions — including multiple agentic AI frameworks, traffic sources, cloud hosting, secured by TEEs, and frictionless on/off-ramps to both fiat and crypto. By allowing developers to seamlessly build and deploy AI, and by sharing revenue with the contributors, the Alliance is paving the way for free AI inference at scale. This united front aims to replace the traditional “pay-to-play” model with a user-first approach for the benefit of all 5.5 billion people connected to the internet.

The following teams are already working together and will share progress in the coming months. For more updates and to join the alliance, you can visit openagentsalliance.org.

The alliance includes: NEAR AI, Coinbase Onramp & AgentKit, Eliza Labs, Aethir, Bitte Protocol, Akash, Phala Network, Hyperbolic, Exabits, SWEAT Economy, HOT, Frax Finance, Arc and MotherDAO

Introducing the Open Agents Alliance was originally published in NEAR Protocol on Medium, where people are continuing the conversation by highlighting and responding to this story.

By Illia Polosukhin

NEAR AI is releasing today a Request for Comments (RFC) of AITP: Agent Interaction and Transaction Protocol. AITP enables AI agents to communicate securely across trust boundaries while providing extensible mechanisms for structured interactions. We anticipate a future where agents are the dominant interface for all online interactions and transactions, representing people, businesses, and institutions and communicating with users and each other.

As AI is becoming more capable in decision-making and taking action, integrating more with existing tools, marketplaces, websites, and apps, it is clear that a lot of existing businesses and applications are going to be replaced by AI Agents. An AI agent is a system that combines reasoning capabilities via LLM and the ability to execute actions in the real world, just as humans are able to do. Three main types of AI agents are:

• Assistants: agent that works on behalf of the user and/or represents the individual
• “Service” agents: agents run by a person or a company for some task, where the company has access to all logs and can change results of the service
• Autonomous agents: represent themselves or their own token holders, require verifiable computation

Agents generally operate using some combination of reasoning/thinking processes (via an AI model) and occasionally following defined rules and workflows. With advances in AI models and supporting infrastructure, AI agents can support experiences that combine the scale and cost of online services with the flexibility and personalization of in-person interactions, leading to net-better user experiences across a range of use cases. But while this future is approaching quickly, making this advancement possible at scale comes with a lot of requirements.

Agents across diverse agent networks–i.e., the internet of agents–need to be able to interact securely and autonomously, make agreements and transactions, and have a common standard for communication. Just as HTTP and HTML enable any web browser to visit any website, AITP provides a standard for agent-to-agent and user-to-agent communication, regardless of where those agents run or how they’re built.

AITP is a big step towards enabling the Internet of Agents: providing a seamless, universal way for agents to interact with people and each other, affording them far more autonomy and functionality and–importantly–the ability to participate in commerce, whether in crypto or fiat. More than adding a new AI layer over today’s web, we are providing a new protocol that fundamentally changes how interactions and transactions take place online. We will see new experiences, new businesses and business models, and new forms of entertainment emerge as the agentic web takes shape.

To read more about AITP–the vision, how it works, and how to use it–and to explore the GitHub repo, visit aitp.dev. The team welcomes feedback and contributions to AITP, which is in the process of integration to NEAR AI Hub (app.near.ai) and various agents built by NEAR AI and our collaborators.

Introducing AITP: Agent Interaction & Transaction Protocol to enable inter-agent payments &… was originally published in NEAR Protocol on Medium, where people are continuing the conversation by highlighting and responding to this story.

The 1980s were a transformative decade that set the stage for the digital world we inhabit today. Cryptography transitioned from academic theory to practical application, personal computing became a household phenomenon, and the open-source movement laid the philosophical and technical foundations for collaborative innovation. These developments not only defined the decade but also catalyzed the rise of cryptocurrencies, artificial intelligence (AI), and the internet.

Cryptography Matures: Building Digital Trust

The 1980s saw cryptographic theory evolve into practical systems that secured the burgeoning digital landscape. Public-key cryptography, first introduced in the 1970s, became widely adopted during this era, enabling secure communication and e-commerce.

One of the decade’s defining moments was the rise of the RSA algorithm, developed in 1977 but gaining traction in the 1980s. RSA was critical for encrypting sensitive information, securing emails, and enabling digital signatures. Its robust security, rooted in the mathematical difficulty of factoring large primes, became a cornerstone of public-key infrastructure (PKI).

Digital signatures were also formalized during this time, enabling tamper-proof authentication of documents and messages. This innovation laid the groundwork for trustless systems like blockchain, where transactions can be verified without centralized intermediaries.

The steps of the RSA Algorithm:

1. Key Generation

The RSA algorithm begins with generating a pair of cryptographic keys: a public key for encryption and a private key for decryption. This process involves the following steps:

1. Select Two Large Prime Numbers (p and q):

• Choose two distinct large prime numbers, ppp and qqq. The security of RSA relies on these primes being sufficiently large and randomly selected.

2. Compute the Modulus (n):

Multiply ppp and qqq to get nnn: n=p×qn = p \times qn=p×q

nnn is used as part of both the public and private keys.

3. Calculate Euler’s Totient (ϕ(n)\phi(n)ϕ(n)):

Compute the totient of nnn, which is the number of integers less than nnn that are relatively prime to nnn: ϕ(n)=(p−1)×(q−1)\phi(n) = (p — 1) \times (q — 1)ϕ(n)=(p−1)×(q−1)

4. Choose the Public Exponent (e):

Select an integer eee such that 1<e<ϕ(n)1 < e < \phi(n)1<e<ϕ(n) and eee is coprime with ϕ(n)\phi(n)ϕ(n). A common choice is e=65537e = 65537e=65537, as it balances efficiency and security.

5. Calculate the Private Exponent (d):

• Compute ddd, the modular multiplicative inverse of eee modulo ϕ(n)\phi(n)ϕ(n): d×emod ϕ(n)=1d \times e \mod \phi(n) = 1d×emodϕ(n)=1
• This ensures that ddd is the unique integer satisfying this equation.

The public key consists of (e,n)(e, n)(e,n), while the private key is (d,n)(d, n)(d,n).

2. Encryption

To encrypt a message mmm (where m<nm < nm<n) using the public key (e,n)(e, n)(e,n):

1. Convert the message into a numeric format mmm (e.g., using ASCII encoding).
2. Apply the encryption formula: c=memod nc = m^e \mod nc=memodn Here, ccc is the resulting ciphertext, which can be safely transmitted over an insecure channel.

3. Decryption

To decrypt the ciphertext ccc using the private key (d,n)(d, n)(d,n):

1. Apply the decryption formula: m=cdmod nm = c^d \mod nm=cdmodn
2. The result mmm is the original plaintext message, which can be converted back into its original form.

4. Digital Signatures

RSA also supports digital signatures, which verify the authenticity of a message:

1. Signing:

• The sender uses their private key (d,n)(d, n)(d,n) to create a signature: s=mdmod ns = m^d \mod ns=mdmodn
• The signature sss is sent along with the message.

2. Verification:

• The receiver uses the sender’s public key (e,n)(e, n)(e,n) to verify the signature: m=semod nm = s^e \mod nm=semodn
• If the computed mmm matches the original message, the signature is valid, confirming the sender’s identity and the message’s integrity.

In 1989, Phil Zimmermann’s creation of Pretty Good Privacy (PGP) brought robust encryption to the masses. PGP’s open-source nature made advanced cryptography accessible to individuals, democratizing security and empowering users to protect their communications in an increasingly digital world. These advancements highlighted the role of cryptographic primitives — key exchange, hashing, and digital signatures — in securing the foundations of modern technologies, from blockchain to crypto wallets.

The Personal Computing Boom: Power to the People

The 1980s were also defined by the explosion of personal computing, which put unprecedented computational power into the hands of everyday users.

The release of the IBM PC in 1981 and the Apple Macintosh in 1984 transformed computing from a specialized tool for scientists and businesses into an essential household and workplace device. These machines were not only more affordable but also user-friendly, enabling widespread adoption.

This democratization of computing power was accompanied by exponential advancements in hardware. Moore’s Law, which predicted the doubling of transistor density roughly every two years, drove innovations in processing speed and memory. These improvements allowed for increasingly sophisticated software, including cryptographic utilities, to run on personal machines.

At the high-performance end, supercomputers like the Cray-2 and the Connection Machine pushed the boundaries of computation, enabling research in parallel processing and distributed systems. These innovations paved the way for future developments in cloud computing and large-scale AI models.

The Rise of Open-Source: A Revolution in Collaboration

The 1980s were a turning point in the philosophy and practice of software development, as the principles of open-source collaboration began to take root. The movement’s origins can be traced back to the early days of computing, when researchers and hobbyists freely shared code to solve problems collectively.

The GNU Project, launched by Richard Stallman in 1983, crystallized this ethos into a formal movement. Stallman envisioned a world where software users had the freedom to run, modify, and share code without restrictions. GNU aimed to create a free operating system, and its principles laid the foundation for the General Public License (GPL), which remains a cornerstone of open-source development.

At the same time, the Berkeley Software Distribution (BSD) played a crucial role in the evolution of open-source operating systems. BSD Unix, developed at the University of California, Berkeley, became a key platform for innovation, particularly in networking. Its contributions to the development of the TCP/IP protocol stack enabled the rise of the internet.

The ethos of openness extended to cryptography, with tools like PGP and cryptographic libraries making secure communication protocols more accessible. This transparency not only accelerated adoption but also fostered trust in cryptographic standards, which are critical for today’s decentralized systems.

Foundations for the Crypto, AI, and Internet Revolution

The innovations of the 1980s in cryptography, personal computing, and open-source software were deeply interconnected, creating the infrastructure for the technological revolutions that followed.

• Cryptocurrency: The cryptographic breakthroughs of the 1980s — especially public-key infrastructure (PKI) and digital signatures — formed the backbone of blockchain technology and cryptocurrencies like Bitcoin, introduced in 2008.
• Artificial Intelligence: While the 1980s were a challenging period for AI, the advances in computing power enabled research into machine learning and neural networks, laying the groundwork for the AI resurgence in the 2000s and 2010s.
• The Internet: The adoption of TCP/IP as the standard networking protocol, rooted in 1980s research, allowed disparate computer systems to communicate seamlessly. Cryptographic protocols like SSL/TLS, developed in the early 1990s, ensured secure data exchange over these networks, enabling e-commerce and online collaboration.
• Open-Source Collaboration: The principles championed by the GNU Project and BSD Unix fostered a culture of transparency and innovation, influencing everything from Linux to blockchain platforms and AI frameworks.

A Legacy of Innovation

The 1980s were a decade of profound transformation, bridging the theoretical and practical realms of technology. Cryptography became a practical tool for securing digital communication, personal computing empowered individuals and businesses, and open-source collaboration revolutionized how software was developed and shared.

As we navigate the interconnected, AI-driven, and decentralized world of the 21st century, the legacy of the 1980s remains unmistakable. The breakthroughs of this pivotal decade continue to shape our technological landscape, inspiring new generations to build on the foundations of trust, accessibility, and collaboration.

About NEARWEEK

NEARWEEK is the ultimate destination for all things related to NEAR. As the official NEAR Protocol newsletter and community platform, NEARWEEK is the one-stop media for everything happening in the NEAR ecosystem.

NEAR Newsletter | Twitter

About NEAR Protocol

NEAR is on a mission to onboard a billion users to the limitless possibilities of Web3 with chain abstraction. Leveraging its high-performance, carbon-neutral protocol, which is swift, secure, and scalable, NEAR offers a common layer for browsing and discovering the Open Web.

NEAR AI | What is Chain Abstraction? | Twitter

THE PATH TO USER OWNED AI #7 was originally published in NEAR Protocol on Medium, where people are continuing the conversation by highlighting and responding to this story.