On June 8, IC3 released a 155-page study focusing on whether artificial intelligence and cryptocurrency can truly complement each other. The findings were cautious: while wallets, blockchains, and decentralized governance can address certain issues, many commonly cited claims within the industry are significantly overstated.
The wallet can only be upgraded automatically.
Researchers believe that crypto wallets can enable AI agents to automatically complete payments, transactions, and service calls, but this does not make the system "autonomous." As long as servers, access permissions, or execution rules are still set and adjusted by humans, the AI remains under human control.
The report states that the primary function of a wallet is to automate payment and on-chain operational processes. AI agents can initiate transfers, exchange assets, or purchase services under predefined conditions without requiring manual approval at each step.
However, the study also emphasizes that these capabilities still depend on external systems. Humans can modify agent rules, shut down the operating environment, or cut off critical services it relies on. In other words, wallets make it easier for AI to perform tasks, but they do not change who holds control.
The study also notes that centralized financial systems can similarly offer programmable payments. The advantages of blockchain networks may lie in neutrality and censorship resistance, but project teams must still demonstrate that on-chain solutions offer measurable benefits over traditional payment tools in terms of cost, accessibility, or stability.
On-chain records do not equate to authenticity verification
Regarding the issue of identifying AI-generated content, IC3 also provides a clearer boundary. The study states that blockchain can timestamp files and store claims about their origin, but the network itself cannot determine whether an image, video, or text was created by a human or a model.
Such judgments still rely on off-chain classifiers or external tools. If the external judgment itself is incorrect, the blockchain will simply preserve the erroneous information intact. Therefore, on-chain records can ensure that the data has not been tampered with, but they cannot guarantee that the originally entered information was accurate.
Decentralized model bias is difficult to resolve
The report also refutes another common claim—that AI becomes fairer simply by moving training, governance, or inference processes onto decentralized networks. The study argues that model bias typically stems from training data, model design, and inference methods, and is not fundamentally solvable by whether or not something is put on-chain.
However, IC3 did not deny the value of cryptographic technologies in the field of AI. The study notes that zero-knowledge proofs, trusted computing, and blockchain can still be used to secure systems, maintain critical records, and enable micropayments between machines. These applications are more specific and closer to real-world implementation.
The study also cautions that directly chaining large datasets, model checkpoints, and inference logs faces cost and scalability limitations. Truly valuable integration scenarios still require more real-world examples to demonstrate their effectiveness.
The product has recently begun pilot testing.
In contrast, several platforms have recently launched wallet or account products designed for AI. MetaMask launched an early test version of its Agent Wallet on June 8, enabling AI to perform on-chain operations such as swaps under user-defined rules. Robinhood has also introduced a dedicated account for agent trading, segregated from the user’s main assets.
Solana and Google Cloud previously launched Pay.sh, attempting to enable AI agents to purchase API services using stablecoins on a pay-per-use basis. IC3 believes such scenarios hold some potential, but the project team still needs to answer a core question: In real-world service environments, is crypto payment truly cheaper, easier to integrate, or more resilient than existing tools?