Anthropic Launches Project Glasswing to Use AI to Find and Fix Critical Software Vulnerabilities

## Executive Summary Anthropic's Project Glasswing leverages the Claude Mythos Preview LLM to autonomously identify and remediate thousands of previously undiscovered vulnerabilities in critical software — notable findings include a 27‑year OpenBSD crash bug, a 16‑year FFmpeg flaw, and chained Linux kernel vulnerabilities enabling privilege escalation — all of which were reported and patched; Anthropic is providing credits and funding to open-source maintainers and restricting access to vetted partners, while observers warn of risks if the model is abused or leaked.