Crypto Exchanges Face New Risk: AI Model That Can Chain Zero-Day Exploits

Security experts warn that a new artificial intelligence model developed by Anthropic could significantly raise the risk of cyberattacks on cryptocurrency exchanges. The AI, called Mythos, is capable of autonomously detecting and chaining together zero-day software vulnerabilities — a capability that has raised alarms in the security community. In one test, the model reportedly identified a decades-old bug in OpenBSD and used it to bypass browser lockouts, a task that typically requires elite security teams. Anthropic launched the model as part of its 'Claude Mythos Preview,' but developers have since described it as 'too dangerous' for public release. In 2025 alone, $3.4 billion in crypto assets were stolen through exploits, and high-profile hacks at exchanges like Bybit and Drift Protocol have underscored the fragility of platform security. Despite these concerns, Coinbase and Binance are exploring the use of Mythos for defensive purposes, with Coinbase’s chief security officer acknowledging that the model will accelerate both digital threats and digital defense. Centralized exchanges, which process billions in user funds and store vast amounts of personal data, are now considered prime targets for AI-powered attacks.