Home / Technology / AI Detects 10,000+ Critical Software Flaws
AI Detects 10,000+ Critical Software Flaws
4 Jun
Summary
- Anthropic's AI, Claude Mythos Preview, identifies over 10,000 severe security flaws.
- Project Glasswing expands to 200 organizations globally, including government bodies.
- AI's potential to exploit software bugs has prompted urgent government and financial sector meetings.
Anthropic is significantly expanding its AI initiative, Project Glasswing, which identifies software vulnerabilities. The program, initially involving about 50 organizations, will now include roughly 200 members that meet specific security requirements to access Anthropic's Claude Mythos Preview AI. This powerful artificial intelligence has already uncovered over 10,000 security flaws deemed highly or critically severe by partners.
The rollout involves collaboration with the U.S. government and other industry players. This expansion follows intense interest from both business and government sectors in Anthropic's AI capabilities, which can reportedly pinpoint ways to exploit small software bugs. This has led to high-level discussions, including a meeting convened by U.S. Treasury Secretary Scott Bessent and then-Federal Reserve Chair Jerome Powell with bank CEOs to address the risks.
Goldman Sachs CEO David Solomon emphasized the need for broader information sharing about these powerful AI models and the implementation of guardrails. The new Glasswing partners are spread across more than 15 countries and bolster defenses in critical sectors like healthcare, power, and communications. Anthropic estimates a major attack on partner organizations could impact over 100 million people.
