OpenZeppelin's CEO Manuel Araoz has issued a stark warning regarding the security of decentralized finance (DeFi News). On the social platform X, he stated that his trust in 'all' DeFi News applications has plummeted, primarily due to the rise of artificial intelligence (AI) agents capable of analyzing code and exploiting system vulnerabilities at a 'superhuman' level.
According to data from DeFi NewsLlama, the total value locked in DeFi News platforms has decreased by $20 billion since the beginning of the year. While part of the decline is related to the overall market downturn, a series of protocol vulnerabilities and concerns over security flaws have significantly heightened investor anxiety.
Data shows that over the past year, the DeFi News ecosystem has lost more than $1.1 billion due to various security incidents. Among these, the Kelp DAO hacking incident in April 2024 resulted in losses of $292 million, exposing serious vulnerabilities faced by cross-chain protocols. Similarly, the Solana-based Step Finance platform was forced to shut down completely after suffering a $27 million hack earlier this year.
Araoz emphasized that a core risk is that attackers are no longer just human; complex AI agents have begun to get involved. One significant development is Claude Mythos, launched by Anthropic, an AI model capable of independently discovering software vulnerabilities and instantly generating effective exploit code. According to Anthropic, the model's performance far exceeds that of current automated security tools.
Users warn that DeFi News, once known for its transparency, is rapidly becoming a burden as advanced machine-driven systems can now exploit this transparency. Since smart contract code is publicly visible, AI can identify flaws faster than humans and launch attacks immediately.
Experts point out that the current security model of DeFi News is built to address threats from human actors, not ultra-fast AI agents. With the arrival of advanced AI agents, risk assessment and defense mechanisms must undergo a complete redesign. These emerging technologies often outpace human response teams, making it difficult for traditional systems to keep up.