[Update] Official Launch: Anthropic Rolls Out Project Glasswing

From closed preview to a full-fledged initiative. On April 12, 2026, Anthropic officially announced `Project Glasswing` on its website, cementing the status of the Claude Mythos Preview model as a key tool for auditing critical software.

If a few days ago it was about local tests, now it is a full-fledged B2B product. Against the backdrop of yesterday's vulnerability in the Axios package, which nearly affected OpenAI, the launch of Glasswing looks perfectly timed. Anthropic is selling corporations not text generation, but an automated AI auditor capable of parsing the codebase for supply chain attacks and architectural holes. The cybersecurity industry is moving from manual penetration testing (pentest) to continuous neural network analysis.

Source: Anthropic