This article is a deep-dive from JudyAI Lab — an AI engineering playbook series with 100+ published guides, 5,000+ weekly readers across 60+ countries, focused on the practical side of running AI agents, trading systems, and content pipelines in production.

📰 Key Takeaways

Anthropic’s release of Claude Mythos (internally codenamed Fable 5) has put the crypto community on high alert. In April, hackers stole a total of $629.7 million in crypto, the highest monthly figure since February 2025, and analysts are linking this directly to the widespread adoption of AI tools.

Moonrock Capital founder Simon Dedic publicly warned that with Fable 5 out in the world, the cost and technical barrier to finding exploitable smart contract vulnerabilities is heading toward zero. He says unaudited DeFi protocols are about to become sitting ducks — known attack techniques will get copied and reapplied across all kinds of forked projects, and even tiny projects won’t escape attention, simply because the cost of attacking them has dropped so low. His advice: revoke wallet approvals proactively, pull assets out of protocols where you can, and move your holdings to a brand-new hardware wallet.

Curve Finance co-founder Michael Egorov takes a more measured view. He notes that the vulnerabilities Mythos has found in other software mostly come from massive codebases running into the millions of lines, while DeFi smart contracts are typically only a few thousand lines — a scope humans and existing AI can already reason through completely, so the breakthrough isn’t as dramatic as it’s being made out to be. He predicts the bigger risk actually sits at the operational security layer — things like stolen multisig keys and supply-chain attacks on frontend dependencies — rather than on-chain contracts getting cracked open directly.

On the capability side, Anthropic ran Project Glasswing in May, having Mythos scan over 1,000 open-source projects, turning up roughly 6,200 high- or critical-severity vulnerabilities. Anthropic has now opened up Claude Mythos 5 to a small batch of security and infrastructure companies, with some guardrails loosened for that purpose.

💬 JudyAI Lab’s Take

The real story here is that AI’s leap in vulnerability-scanning capability is flipping the entire cost equation behind DeFi’s security defenses.

Anthropic’s Project Glasswing had Claude Mythos scan over 1,000 open-source projects and turn up roughly 6,200 high-severity vulnerabilities — a clear sign that AI as a dual-use offense/defense tool is scaling up fast. Moonrock Capital founder Simon Dedic’s warning points to a real problem: once the technical barrier to finding contract vulnerabilities drops toward zero, even tiny, unaudited protocols won’t escape attention. That said, Curve Finance co-founder Michael Egorov offers a useful counterpoint — smart contract codebases are limited in scope, so the breakthrough AI brings may not be as dramatic as it’s being portrayed; the bigger practical risk actually comes from operational-layer issues like stolen multisig keys and supply-chain attacks on frontend dependencies. That’s a good reminder that facing AI-driven security threats, you shouldn’t just be staring at the on-chain contracts — layered defense across your whole operating environment matters just as much.

Take action now: go into every DeFi protocol you have money in and revoke wallet approvals you’re no longer using, and think about whether it’s time to move your core holdings to a hardware wallet — right now, that’s the cheapest form of self-protection available.

📅 Source Info

🔗 Further Reading

References