AI Speed Threats: Crypto’s New Hacker Tool

Anthropic has launched its most powerful AI model yet, Claude Fable 5, which can think and code faster than previous versions. The release comes as decentralized finance (DeFi) has already lost more than $840 million to hacks this year. The new model is split into a public version and a restricted one called Claude Mythos 5, intended only for vetted security teams. The public model is designed to block dangerous requests by diverting high‑risk queries to a weaker system, Claude Opus 4. 8. Anthropic claims this fallback happens in fewer than 5 % of sessions and that extensive testing, including over a thousand hours of bug‑bounty work, found no way to break the safety net. Yet the company admits that motivated attackers could still try to defeat these measures because the power of the model is highly valuable. Experts point out that AI’s biggest advantage for a hacker is not finding new kinds of exploits but doing the work in seconds. A reasoning model can automatically scan code, compare versions, and spot misconfigurations at machine speed. In crypto, where software bugs can translate instantly into financial loss, this speed gives attackers a huge edge.

Recent big losses in DeFi have not been due to smart‑contract bugs that AI could engineer. Instead, they stem from social engineering, poor key management, and human error. For example, a North Korea‑linked group stole $285 million from Drift Protocol after six months of social engineering, and Kelp DAO lost $292 million through a single‑verifier flaw. Even a private‑key compromise cost Humanity Protocol over $30 million when an employee’s laptop was breached. AI can aid attackers by reading public code, summarizing audit reports, and drafting convincing phishing messages. It can also help defenders by mapping codebases, stress‑testing contracts, and catching bugs early, as Pendle has done. However, the real protection comes from secure key storage and trusted signing displays that keep private keys out of reach from compromised devices. In short, the next major crypto hack will likely look similar to past incidents—social engineering or bad signing flows—but it could happen faster thanks to AI. The industry must focus on hardware security and clear signing practices to keep pace with the speed of machine‑generated threats.

Actions