Anthropic is finding bugs faster than Microsoft can fix them
Ars Technica AI · Renee Dudley, ProPublica · 2026-07-29
Anthropic's AI model Mythos, shared with select organizations under Project Glasswing, is discovering software vulnerabilities in Microsoft's codebase faster than engineers can patch them, raising urgency about defensive AI parity with nation-state adversaries.
Appears in
Extraction
Topics: ai-securityvulnerability-researchanthropicmicrosoftoffensive-ai
Claims
- Anthropic developed an AI model called Mythos that discovers software vulnerabilities at an unprecedented rate compared to human-led security research.
- Microsoft convened emergency cross-team engineering meetings under Project Glasswing to address vulnerabilities Mythos uncovered faster than their teams could fix them.
- Anthropic gave select software organizations access to Mythos with the explicit goal of finding and fixing vulnerabilities before adversarial governments like China could exploit similar AI tools.
- The deployment of Mythos raised internal debate at Microsoft about whether the AI lived up to Anthropic's claimed capabilities.
Key quotes
The tech giant was racing to fix weaknesses in its code that a new AI model known as Mythos was uncovering at an unprecedented clip.
one engineer asked the question that loomed over the meeting: Did Mythos 'live up to the hype that Anthropic claimed it would have had?'
The goal was to find and fix the vulnerabilities before hackers and adversarial governments like China began using similar tools to find and exploit them for espionage and sabotage.