Anthropic's AI Model Poses Threat to DeFi Security

On June 13, 2026, Anthropic launched its Claude Fable 5 AI model, the first in the Mythos class, which offers enhanced reasoning and coding capabilities. One version is accessible to the public, while a more powerful variant is restricted to vetted cybersecurity users. This advanced technology arrives amidst significant security challenges in the decentralized finance (DeFi) sector, which has faced over $840 million in hacks in the first five months of the year.

Experts contend that while AI like Mythos can find previously unknown software flaws, the real concern lies in its ability to exploit existing weaknesses much faster. Most of this year's DeFi losses stem from human errors and operational failures, rather than vulnerabilities within smart contracts themselves. Notably, one of the largest breaches involved a North Korea-linked hacker group siphoning approximately $285 million from the Drift Protocol through a months-long social engineering campaign.

Despite Anthropic's safeguards, such as rerouting high-risk requests, experts warn that determined and financially motivated attackers may still find ways to bypass these controls. Charles Guillemet, Chief Technology Officer at Ledger, remarked on the potential speed advantage AI brings, noting that, "Current AI guardrails raise friction. They are not a reliable control against a determined adversary." He cautioned that the efficiency with which AI can analyze code and configurations could accelerate the pace at which exploits are developed.

Data from DefiLlama shows that April alone saw $600 million lost to hacks, marking it as a particularly disastrous month for the industry. The nature of these incidents underscores the fact that many exploit pathways stem from human missteps rather than technical flaws. This reality has resulted in security protocols needing to secure each operational path rigorously and maintain high standards of integrity.

To mitigate risks, projects like Pendle, a DeFi yield protocol, have begun leveraging Anthropic’s AI defensively to enhance code quality and identify vulnerabilities early in the development process. They suggest that while the threats may not be new, the presence of AI could lead to faster exploitation of familiar vulnerabilities, increasing their urgency.

Ultimately, the integration of advanced AI into the hacking landscape poses a dual challenge: on one hand, it can improve security measures; on the other hand, it accelerates the discovery and exploitation of weaknesses, making rapid responses essential in the DeFi space.