AI agents speeding up crypto fraud, analysts warn

The speed at which AI agents operate allows traditional cryptoasset crime models to function faster while also introducing new attack surfaces and forcing the development of countermeasures.

Cryptoasset industry players have warned that AI agents compress financial crime timelines and break fraud's cost curve.

For example, blockchain analysis company TRM Labs noted that while the speed of post-compromise fund movement is crucial for investigation and recovery efforts, the fact that autonomous agents are becoming more common in treasury management, trading, and liquidity operations makes this even more complicated.

"If compromised or misconfigured, an AI-driven wallet manager could fragment funds across dozens of addresses, convert assets through multiple liquidity pools, and route value across blockchains before a human operator becomes aware of anomalous activity," the analysts said.

Moreover, according to them, these potential threat agents introduce new control dependencies and technical attack surfaces.

For example, agents holding signing authority over treasury assets or operational wallets might become high-value targets for prompt injection, adversarial data manipulation, compromised governance keys, or attempts to trigger unauthorized transfers.

Also, criminals can launch agents designed to automate laundering workflows, exploit decentralized protocol vulnerabilities, or dynamically adjust transaction routing to avoid known detection patterns.

The third new risk is related to the possible misaligned optimization of an agent. This could include routing funds through high-risk liquidity venues or interacting indirectly with sanctioned infrastructure just to maximize yield or efficiency.

Meanwhile, in his detailed post this week, Daniel Barabander, investment partner and chief legal officer at crypto venture capital fund Variant, also emphasized that crypto and AI dramatically improve fraud's unit economics and that fraud is now a profit-driven business.

As a countermeasure, Barabander suggested using so-called cryptographic proofs of reality, or physical-world signals AI can't easily spoof, such as device-signed location. This could enable use cases like geofenced remittances, wire approvals, and account-takeover blocks.

Meanwhile, recovering stolen funds might be aided by "crypto kill chain" networks designed for quick tracing, freezing, and seizure of stolen assets.

"If the fraud thesis is correct, the next decade will not simply see more fraud. It will see the construction of a new security layer designed to contain it," the investment partner concluded.

In other words, as TRM Labs summarized, AI-enabled financial crime risk requires an AI-enabled compliance and investigative response.

Unlock more exclusive Cybernews content on YouTube.