Researchers from major U.S. universities said industry discourse around combining artificial intelligence (AI) and blockchain includes exaggeration, and outlined 5 representative misconceptions.

On June 9, blockchain outlet CoinPost reported that IC3, an academic research consortium based at Cornell University, released a paper titled "Crypto x AI, AI x Crypto: A Survey". The paper involved 25 researchers from several institutions including Cornell Tech, Carnegie Mellon University, Princeton University and Yale University.

The researchers said discussion about the intersection of AI and blockchain has grown rapidly since the spread of generative AI, but real opportunities and limits are not being sufficiently distinguished. They said combining AI and blockchain could create new possibilities, but blockchain or token structures alone cannot automatically solve AI reliability, fairness and cost problems.

The paper said it is appropriate to view the relationship between AI and blockchain as middleware that complements each other. It said AI could make blockchain a more flexible and easier-to-use system, while cryptographic technologies could strengthen security and governance for AI systems.

It cited on-chain fraud detection, identifying malicious smart contracts and analysing code vulnerabilities as examples of how AI could be used in the blockchain domain. It added that many of these approaches are based on relatively simple machine-learning models and are more effective when sufficient training data are available.

It also set out areas where blockchain and cryptographic technologies could complement AI. It said technologies such as zero-knowledge proofs and trusted execution environments (TEE) could be used to improve security and verifiability of AI systems. It added that methods experimented with by cryptocurrency communities, such as decentralised governance and infrastructure operations, have not yet spread widely into the mainstream AI field.

The paper placed particular emphasis on 5 misconceptions spreading in the industry. The first is the claim that blockchain can detect AI-generated content itself. The researchers said blockchain can verify metadata for authenticity checks, but cannot determine whether content was generated by AI based on content alone.

The second is the claim that decentralisation technologies solve AI bias and fairness problems. The researchers said decentralised governance can increase transparency in decision-making, but cannot eliminate bias embedded inside algorithms.

The third is the claim that giving AI agents wallets makes them autonomous entities that earn and spend money on their own. The researchers drew a line between automation and autonomy, saying they are different concepts. They said adding payment functions does not make an AI agent an independent economic actor, and blockchain is not necessarily required for payment automation.

The fourth is the claim that recording training data or inference results on a blockchain enables trustworthy AI operations. The researchers said blockchain can prevent post hoc tampering of data, but cannot guarantee the reliability of the original data itself. They added that blockchain throughput and costs could also be obstacles to verifying large-scale AI processes.

The fifth is the claim that decentralisation inevitably lowers the costs of AI training and inference. The researchers said a decentralised structure could be more expensive than a centralised approach depending on network latency and throughput conditions. They said the claim that combining AI and cryptocurrencies directly leads to cost reductions still needs verification.

As a future task, the paper said AI safety should be addressed from the perspective of the overall system rather than individual models. It said the AI industry currently focuses on model-level responses such as guardrails that control inputs and outputs, but this may be insufficient if AI agents gain broader access to financial systems and infrastructure.