AI and crypto are still mostly early-stage hype, study says

The symbiosis of crypto and AI may be everywhere in pitch decks, token launches and agent-wallet demos, but a new 155-page research survey says the actual merger of the two technologies is still in its early innings.

The June 8 paper, published by IC3, the Initiative for CryptoCurrencies and Contracts, edited by Giulia Fanti and Ari Juels and written by researchers from Cornell Tech, Carnegie Mellon University, Princeton University and Yale University, shows that the AI and crypto boom still has more questions than proof behind it.

The authors said the intersection of crypto and AI is "spawning papers, products, online posts, and companies," but warned that the surrounding buzz can hide what has actually been built and what still needs proof.

While the paper doesn't dismiss the idea that AI and crypto can work together, it points to a narrower and more uncomfortable problem for the sector. The strongest use cases still need more evidence.

Why the crypto layer matters

As the authors argue, AI and crypto are "still in the very early stages of meaningful integration," even as the market fills with agent wallets, decentralized AI networks, tokenized compute pitches and payment tools for autonomous software.

• Some of those ideas may prove useful, but the report keeps circling back to the same test. The crypto layer still has to explain why it needs to be there.

That question is clearest in decentralized AI, where projects are trying to move training or compute away from the big cloud platforms.

But the paper points out that these efforts still need "more rigorous and direct cost comparisons" with centralized AI platforms, because showing that decentralized training can work doesn't yet show that it can beat cloud giants on cost, speed, reliability or access.

Read also: X402 founder leaves Coinbase to start new company

Second chance after an old failure

Internet micropayments have been tried before. The report notes that HTTP even included a "Payment Required" status code for a web where users could pay tiny amounts for content or services on the fly.

And crypto didn't fix that on its own. Although cryptocurrencies lowered some payment costs and removed some traditional intermediaries, the core problem remained unchanged, the report notes.

Wallets ≠ autonomy

The same tension runs through AI payments. Crypto rails could make sense for agents that need to move value across platforms, especially where traditional payments are slow or closed.

But the report says the sector still has to show measurable benefits over centralized payment systems, rather than only proving that an agent can technically hold a wallet and send money on-chain.

That is where the paper cuts into one of the easiest marketing lines in crypto AI.

Giving an AI agent a wallet may let it trade, pay for services or interact with smart contracts without a human approving every step, but the authors warn that this should not be confused with independence.

Risks may be easier to prove than the business case

The study's more convincing case for crypto isn't that every AI agent needs a token or wallet. As the authors claim, AI systems may need stronger ways to prove what happened.

• For instance, tools such as zero-knowledge proofs, verified execution or authenticated AI pipelines could help prove what an AI system did, where the data came from or whether a model actually ran as claimed.

Still, the risk side is harder to shrug off. The authors warn that combining AI agents with crypto could create "new threat actors and vectors," including rogue smart contracts and "unstoppable autonomous agents."

The tools may matter, especially around verification and payments. But the industry still has to prove that crypto adds something users can't already get from existing AI platforms or payments solutions.

Read more: Mastercard taps Coinbase, OKX and Aave for AI payments