OpenZeppelin chief warns AI makes DeFi unsafe

TLDR

OpenZeppelin co-founder Manuel Aroz has said he now considers all of DeFi unsafe because AI-powered tools are making it much easier for attackers to find and exploit weaknesses.

1. Aroz publicly urged friends and family to exit all DeFi, citing superhuman AI coding agents and a growing asymmetry between attackers and defenders.
2. His warning comes after over $1 billion was stolen from DeFi in recent months, with AI helping automate exploit discovery, especially in cross-chain and admin-key failures.
3. For users, this does not mean every protocol will fail tomorrow, but DeFi should be treated as a high-risk bucket, with extra scrutiny on security design and operational practices.

Deep Dive

1. What The OpenZeppelin Chief Actually Said

In a recent X post, Manuel Aroz said he now considers all of DeFi unsafe and has advised friends and family to exit DeFi positions, including blue-chip protocols like Aave, MakerDAO, and Compound. Multiple outlets report that he believes AI coding agents are now superhuman at finding vulnerabilities, creating an asymmetric fight where defenders must fix every bug while attackers need only one exploit to steal funds. His comments are covered in detail by both Crypto.news and The Block via TradingView.

2. How AI Is Changing DeFi Attack Surface

According to a CoinDesk analysis, DeFi has lost over $1.1 billion to hacks in the last year, while total value locked has dropped by more than $20 billion. April alone saw roughly $600 million stolen across exploits like KelpDAO and Drift, making it one of the worst months on record, with North Korea-linked groups behind a large share of the damage, as detailed by Decrypt. Researchers and security firms say AI is automating vulnerability scanning and reconnaissance on open-source smart contracts and infrastructure, allowing attackers to scale their efforts dramatically while also powering malware campaigns that target developer tools and AI assistants themselves, as seen in TrapDoor-style attacks described by CryptoSlate.

3. What Users Should Infer, Not Panic-Buy Or Panic-Sell

Arozs view is deliberately conservative: treat DeFi as if the security model has shifted from hard but manageable to systemically fragile under AI-accelerated attacks. Other experts emphasize that defenders can also use AI for continuous auditing and monitoring, and that many big incidents still come from operational failures like compromised admin keys or bridges, not just contract bugs. Practically, this suggests treating DeFi as a high-risk allocation: prioritize protocols with strong operational security (multisig, timelocks, minimized admin powers), be especially wary of cross-chain and very high-yield schemes, and assume that older, unaudited, or unmaintained contracts are prime AI targets.

Confidence: high because multiple independent reports quote Aroz directly and recent exploit statistics across sources are consistent.

Conclusion

The OpenZeppelin chiefs warning reflects a real shift in the risk math for DeFi as AI supercharges attackers, particularly against transparent, composable smart contracts and fragile infrastructure like bridges and admin systems. DeFi is not guaranteed to implode, but the bar for acceptable risk has moved up, and users who want on-chain yield or leverage should approach it as an experimental, high-risk layer that demands much stricter protocol selection and ongoing security awareness.