Tech giants launch AI-powered ‘Project Glasswing’ to identify critical software vulnerabilities

Anthropic launched Project Glasswing, partnering with major tech firms to use its unreleased Claude Mythos Preview AI model to find and help patch thousands of previously unknown security flaws in critical and open-source software (examples include decades-old bugs in OpenBSD and FFmpeg). The company pledges funding and credits to support the effort, requires partner sharing of findings, and frames the initiative as a defensive response to AI’s dual-use risks while engaging with government stakeholders.