Security Expert Warns AI Advances Make DeFi Protocols Unsafe

Manuel Aráoz, a former CTO and co-founder of OpenZeppelin, has publicly declared that he now considers all decentralized finance (DeFi) protocols to be unsafe [1]. Aráoz argues that the emergence of "superhuman" AI coding agents has fundamentally shifted the security landscape, making it increasingly difficult for human defenders to protect smart contracts against automated exploits [1].

Key takeaways

• Manuel Aráoz, who departed OpenZeppelin in 2019, has advised friends and family to exit all DeFi positions due to rising AI-driven security risks [1, 2].
• Over $1.1 billion has been lost to DeFi hacks in the past 12 months, contributing to a decline of more than $20 billion in total value locked this year [1].
• Anthropic’s restricted Claude Mythos model has demonstrated the ability to autonomously discover and weaponize software vulnerabilities [1, 2].
• Industry leaders, including the founders of Aave and Uniswap, argue that AI can also be used by defenders to strengthen protocols and improve resilience [2].

The Rise of AI-Driven Vulnerabilities

The core of Aráoz’s warning lies in the asymmetry of smart contract security: defenders must identify and patch every potential bug, while an attacker only needs to find one exploit to drain funds [1, 2]. This challenge has been exacerbated by the release of advanced AI models like Anthropic’s Claude Mythos, which can autonomously scan code and develop working exploits [1]. Because DeFi protocols rely on transparent, on-chain code, these machine systems can analyze and weaponize weaknesses at speeds that exceed human capabilities [1].

The impact of these vulnerabilities is already visible in the market. Recent incidents, such as the $292 million Kelp DAO exploit in April and the $13.5 million compromise of stablecoin issuer StablR, have highlighted the fragility of current systems [1, 2]. While some hacks involve complex smart contract bugs, others stem from centralized points of failure, such as poor operational security or the misuse of administrative privileges in multisignature wallets [2].

Industry Response and Future Outlook

The reaction to Aráoz’s assessment has been polarized. OpenZeppelin issued a statement clarifying that Aráoz’s views do not represent the company’s current position, emphasizing that they remain committed to securing on-chain finance through AI-augmented defense [1]. Other industry figures, such as Aave founder Stani Kulechov and Uniswap founder Hayden Adams, argue that AI is not a net negative for the industry [2]. They contend that security researchers and whitehat hackers are already utilizing the same AI tools to proactively harden protocols [2].

Some developers suggest that the industry must adopt more rigorous safeguards to survive this new environment. Uttam Singh of Alchemy has proposed the implementation of circuit breakers, timelocks, and security councils to mitigate risk until the sector matures [2]. Meanwhile, initiatives like Block’s Project Loupe are attempting to flip the script by using AI to scan open-source software for vulnerabilities, providing defenders with the same high-level tools currently available to potential attackers [2].