Anthropic Gives Dangerous AI Model to 12 Tech Giants
Anthropic launched Project Glasswing, giving its powerful AI model Mythos Preview to 12 major tech companies to find zero-day vulnerabilities in critical infrastructure. The model outperforms existing tools by 25% in automated vulnerability detection.
Why it matters
Mythos Preview represents a major leap in AI's ability to find serious vulnerabilities in widely-used software, outperforming all existing security tools. Anthropic's restricted access model aims to get these bugs fixed before bad actors can exploit them.
Key Points
- 1Anthropic gave its AI model Mythos Preview to 12 tech giants including AWS, Apple, Google, and Microsoft
- 2Mythos Preview can reason about code like a senior security researcher, finding vulnerabilities that evaded human and automated detection
- 3It discovered a 27-year-old flaw in OpenBSD and a 16-year-old bug in FFmpeg that 5 million tests missed
- 4Anthropic restricted access to Mythos Preview to responsibly disclose and patch vulnerabilities before attackers can exploit them
Details
Anthropic's Mythos Preview is an AI model that can read code, reason about potential vulnerabilities, construct proof-of-concept exploits, and verify bugs - capabilities that surpass existing automated security tools. Benchmarks show Mythos Preview outperforming the previous best model by 25% on vulnerability detection. The model has already found critical zero-day flaws in widely-used software like OpenBSD and FFmpeg that evaded human and automated detection for decades. Anthropic is restricting access to Mythos Preview, giving it only to 12 major tech companies that maintain critical infrastructure, in order to responsibly disclose and patch vulnerabilities before they can be exploited. The company has also committed $100 million in compute credits and $4 million to open-source security foundations to support this effort.
No comments yet
Be the first to comment