Anthropic launches Project Glasswing with Claude Mythos Preview model
Anthropic unveiled Project Glasswing, a security initiative powered by a new frontier model, Claude Mythos Preview, aimed at finding software vulnerabilities.
Anthropic has announced Project Glasswing, an initiative the company describes as urgent work to help secure the world's most critical software. The effort is powered by a newly disclosed frontier model, Claude Mythos Preview, which Anthropic says can find software vulnerabilities better than all but the most skilled human researchers.
The announcement, made via Anthropic's official account on X, positions vulnerability discovery as a headline capability for the new Mythos Preview model — a notable framing shift for a frontier lab, which have historically led model launches with reasoning, coding, or agentic benchmarks rather than offensive security performance. For organizations that build, buy, or operate AI systems, the move signals that model-assisted code auditing is moving from research demos toward dedicated vendor programs, and raises familiar dual-use questions about how access to such capabilities will be gated. Anthropic has not yet published details on availability, pricing, or how Glasswing will engage with software maintainers.