OpenZeppelin co-founder Manuel Aráoz has delivered one of the strongest warnings yet from a major crypto security veteran, arguing that AI coding agents are making decentralized finance harder to defend against sophisticated exploit discovery.
Aráoz’s May 26 post was blunt: “all of DeFi is unsafe.” He tied that view to the asymmetry at the center of smart contract security. Defenders must find and patch every meaningful vulnerability before funds are exposed. Attackers need only one exploitable path, and AI-assisted code review is making that search faster, cheaper and more scalable.
He also said he has personally advised friends and family to exit DeFi positions. That message landed heavily because OpenZeppelin is not a fringe security name. Its audit archive includes work across major DeFi and crypto infrastructure, including Aave, Uniswap, Compound and Coinbase. OpenZeppelin’s newer security program says the firm has conducted more than 900 audits and identified over 10,000 vulnerabilities.
AI Turns Smart Contract Security Into A Speed Problem
The warning reflects a deeper shift in DeFi risk. Smart contract audits were built for a world where teams could review code, patch known issues, launch a protocol and rely on bug bounties or monitoring to catch the rest. AI agents change that timing. They can scan contracts, simulate attack paths and compare patterns across large codebases at a pace human auditors cannot match.
That does not mean every DeFi protocol is actively compromised. It means the defense model is under pressure. Large protocols may still have audits, formal verification, circuit breakers, multisig controls and monitoring, but capital is exposed across smart contracts, governance systems, oracle feeds, bridges, front ends, private keys and third-party integrations.
OpenZeppelin’s own four-layer DeFi risk framework separates those threats into smart contract bugs, key-management failures, governance and upgrade attacks, and cross-chain or dependency exploits. That framework fits recent losses where the failure was not always a simple coding bug.
The backdrop is already severe. DeFiLlama’s hack database tracks more than $7.7 billion in DeFi losses, and recent incidents have kept that risk in view. The $292 million KelpDAO event showed how bridge and integration failures can ripple through lending markets, while CryptoAdventure’s coverage of DeFi exploit losses reaching $816.9 million in 2026and Ethereum smart contract attacks topping $1.5 million in two days shows how broad the attack surface has become.
For users, the practical risk is exposure size. DeFi still offers liquidity, borrowing, trading and yield tools that centralized platforms cannot fully replicate, but the security assumptions have changed. Larger positions now need stronger controls: protocol diversification, limited approvals, hardware wallets, active monitoring, withdrawal plans and a clear reason for accepting smart contract, oracle, bridge and governance risk.
