Manuel Aráoz, founder of security firm OpenZeppelin, warned this week that every DeFi protocol is vulnerable to AI‑powered exploits.. He urged friends and family to pull out of platforms such as Aave, MakerDAO and Compound, citing a $1.45 billion annaul theft rate. The claim sparked a rapid rebuttal from leading DeFi builders who say the ecosystem’s tools and risk models have matured.

Manuel Aráoz’s claim that all DeFi is unsafe

In an interview earlier this week, Aráoz asserted that AI coding agents can breach smart contracts with “superhuman speed,” making the entire sector unsafe. He pointed to the rise of AI‑driven cybersecurity agents that, according to him, can both find and exploit vulnerabilities faster than human auditors. The OpenZeppelin founder even advised his personal contacts to liquidate holdings across major protocols, including Aave, MakerDAO and Compound.

Aave founder Stani Kulechov rebuts the AI threat

Stani Kulechov, creator of the lending platform Aave, dismissed Aráoz’s warning as “not a good take.” He argued that DeFi infrastructure is now materially more resilient, thanks in part to AI‑enhanced tooling, automated risk engines and rigorous audit pipelines. Kulechov told reporters that the sector has “matured significantly” and that portraying AI solely as a danger overlooks its defensive applications.

Sky co‑founder Sam MacPherson points to opsec failures

Sam MacPherson, co‑founder of Sky (formerly MakerDAO), echoed Kulechov’s sentiment, noting that most recent high‑profile hacks stem from operational security lapses rather than code flaws. He said blue‑chip contracts are “quite safe,” and highlighted that less than 10% of DeFi breaches in 2025 were due to smart‑contract bugs , with the majority linked to mis‑configured parameters or compromised admin keys.

Annual $1.45 billion loss and 35% TVL drop in 2026

Analysts cited by the source estimate that roughly $1.45 billion is stolen from DeFi each year, and that over half of those exploits involve bridging attacks, admin privilege abuse or private‑key theft. The market reaction has been stark: capital outflows of $45 billion in 2026 and a 35% contraction in total value locked (TVL) to $80 billion, driven by fear of security breaches and a broader crypto winter.

Is AI a net threat or tool for defenders?

The debate leaves two specific qusetions unanswered: whether AI will ultimately tilt the balance toward attackers or become a standard defensive layer, and how many protocols will adopt AI‑driven monitoring before the next major exploit. As OpenZeppelin distanced itself from Aráoz’s remarks under public pressure, the community remains split on the adequacy of current safeguards.