Anthropic Unveils Claude Mythos Preview With Powerful Zero-Day Detection Capabilities

**Anthropic's Claude Mythos Preview** is described as an advanced language model capable of autonomously finding and chaining software vulnerabilities into fully working exploits; internal tests reportedly achieved control-flow hijacking on patched targets and discovered critical bugs in OpenBSD and FFmpeg. The company launched Project Glasswing to limit access to trusted defenders and use the model defensively, while acknowledging the potential for misuse if such exploit-generation capability were to become broadly available.