Might AI agents uncover the next wave of hidden DeFi exploits?

crypthub

about 19 hours ago

AI-Driven Vulnerabilities

Manuel Aráoz of OpenZeppelin says AI agents can rapidly find exploit loops, making all DeFi “unsafe.” Attackers need only one bug, while defenders must patch every flaw, creating a severe security asymmetry. Coding agents now operate at superhuman speed, threatening even large protocols.

April Hacks and the AI Debate

April saw record crypto hacks, with Dune Analytics reporting $285 M linked to DPRK groups and $437.4 M from unknown actors, mainly via bridge verification flaws and social engineering. While AI accelerates bug discovery, analysts note human errors—faulty signatures or exposed keys—still drive many exploits. The founder of Slow Mist advises DeFi teams to run AI‑powered attack drills each quarter.

May Slowdown and TVL Outlook

In May, attacks dropped to roughly $44 M across 14 incidents, ThorChain being the most serious. Total Value Locked fell to $81 B from over $98 B in April, reflecting lower ETH prices and withdrawals; Aave now holds about $14 B. The slowdown suggests a temporary reprieve, but vigilance remains essential.