A consortium of 25 computer science researchers concludes in a 155-page survey that cryptocurrencies offer only limited value for AI's biggest trust and payment problems. This work by IC3 on AI and blockchain thereby tempers a dominant market narrative.
IC3, the Initiative for CryptoCurrencies and Contracts, is an academic research consortium founded in 2015 that conducts fundamental research on crypto protocols and smart contract security. Thirteen universities participate, among them Cornell Tech, Carnegie Mellon, Princeton, ETH Zurich and the University of Bern. The paper "Crypto x AI, AI x Crypto: A Survey" appeared in June 2026, with Giulia Fanti (Carnegie Mellon University) and Ari Juels (Cornell Tech) serving as co-editors. Juels is also Chief Scientist at Chainlink Labs and co-author of the 2017 Chainlink whitepaper. In total, 25 authors from academia and industry contributed, including researchers from Flashbots, Offchain Labs, Ava Labs and Ritual Labs.
Subscribe to our newsletter
The best articles of the week, directly delivered into your mailbox.
Three theses on AI and blockchain the paper rejects
The first refuted promise concerns the autonomy of AI agents. Advocates argue that a crypto wallet makes an agent autonomous because it controls funds on its own. The researchers disagree: wallet access does enable automation, meaning programmatic action without human approval loops, yet not genuine autonomy. AI agents therefore remain dependent on humans and the underlying infrastructure. A system does not become more intelligent simply because it holds a wallet, nor more resilient against human manipulation or shutdown. Consequently, traditional payment systems could deliver the same automation.
"Automation must not be confused with autonomy." - IC3 researchers, Crypto x AI, AI x Crypto: A Survey
The second point concerns the detection of AI-generated content. Blockchains work as high-integrity timestamping and registration systems, however not as a classifier that distinguishes human content from machine content. An external classifier therefore remains necessary. If it makes an error and the result lands on-chain, the blockchain preserves that error permanently instead of correcting it.
The third point concerns algorithmic biases that arise during training and can be mitigated only through revised training or inference techniques. Decentralization therefore does not address the root cause: it does increase the transparency of governance, yet robust evidence for an actual reduction in bias is lacking. Ultimately, all three theses rest on the same pattern: blockchain shifts the trust problem instead of solving it.
Where blockchain genuinely helps AI according to IC3
The paper passes no blanket anti-crypto verdict, but instead delineates the plausible application areas precisely. As a promising mechanism, the researchers cite zero-knowledge proofs, which allow AI outputs to be verified without disclosing model internals. Furthermore, the work points to trusted execution environments, which make AI computations verifiable within a secured hardware setting. Together, both methods form a cryptographic verification layer for AI services that makes results checkable without exposing sensitive models.
In addition, the paper sees value in provenance records: blockchain timestamps make content and data pipelines for AI training more traceable. Programmable, censorship-resistant stablecoin payments for per-request API calls likewise count as legitimate, for which the authors cite the Solana-Google Cloud product Pay.sh. Such payments additionally require no per-agent KYC and run around the clock, meaning outside banking hours. The decentralization of training and inference is classified as potentially useful by the researchers, though they call for stronger quantitative evidence.
In the reverse direction, the survey acknowledges that AI can in turn strengthen blockchains, for instance in fraud detection, smart contract security and real-time data processing. At the same time, it warns that AI-driven trading systems enable new forms of collusion and market abuse. Ongoing IC3 projects such as Self-Sovereign Agent, IMMACULATE or OML finally mark which of these questions remain open. Self-Sovereign Agent examines when a crypto wallet makes sense for an agent at all, IMMACULATE checks model substitution and token over-charging, while OML explores primitives for a monetizable AI model distribution.
Agentic payments: what the market is building in parallel
While the researchers urge caution, the industry has long been building the infrastructure for agentic payments. One of the most prominent ongoing protocols is x402, which Coinbase and Cloudflare launched in May 2025. It uses the HTTP status code 402 ("Payment Required") to embed stablecoin payments directly into HTTP interactions. First, the server responds to an agent request with the code 402 and payment instructions, whereupon the agent constructs a signed payment and settles it on-chain. The transactions run gasless via EIP-3009 and Permit2. Finality on Base with USDC takes roughly two seconds, and on Solana only around 400 milliseconds. The protocol is open-source and charges no fees of its own. Behind the x402 Foundation, founded in September 2025, stand Google, Visa, AWS, Circle, Anthropic and Vercel, in addition to Coinbase and Cloudflare.
Meanwhile, further providers are pushing into the market. Consensys launched the self-custodial MetaMask Agent Wallet in June 2026 as an early-access program with 200 slots. It lets AI agents act autonomously on EVM chains and Hyperliquid, secured by daily spending limits, allowlists and transaction protection of up to 10,000 USD per month. The startup Skyfire additionally combines its USDC rail KYAPay with an extensible identity token that links platform, agent and human identities. Moreover, it raised 9.5 million USD from Coinbase Ventures, a16z CSX and Neuberger Berman, while partners such as Anthropic, Cohere and Hugging Face are testing the usage-based billing.
Robinhood, in turn, opened its agentic trading in May 2026 to 27 million funded accounts, initially only for stocks, with crypto trading announced as the next step. The paper's central demand targets precisely this wave: advocates must demonstrate that crypto actually performs better than existing payment methods on cost, access or resilience.
