Anthropic's Mythos AI Used to Discover Critical macOS Vulnerabilities
Security researchers from Palo Alto-based firm Calif have used Anthropic's advanced Mythos AI model to discover two previously undocumented vulnerabilities in Apple's macOS operating system. The bugs, when chained together, could allow privilege escalation — potentially letting an attacker seize control of a computer when combined with other exploits.
The researchers delivered a 55-page report to Apple in person at their Cupertino headquarters. Apple confirmed it is reviewing the findings and stated that security remains its top priority. Calif's CEO Thai Duong emphasized that the attack "couldn't have been pulled off by Mythos alone" and required significant human cybersecurity expertise alongside the AI assistance.
The exploit development took only five days with Mythos's help, compared to what would typically take weeks or months. Mythos was also reportedly able to identify over 2,000 hidden software vulnerabilities in a seven-week testing period, more than 30% of which were previously unknown. This case represents one of the most significant real-world demonstrations of AI's dual-use potential in cybersecurity — both as a defensive tool for finding bugs before malicious actors do, and as a potential offensive weapon if misused.
This is a watershed moment for AI-assisted security research. Mythos compressing exploit development from weeks to five days shows AI is fundamentally changing the economics of both offense and defense in cybersecurity.
What is Anthropic's Mythos AI?
Mythos is Anthropic's most powerful AI model designed specifically for advanced reasoning tasks, including cybersecurity research. It has been kept largely restricted due to its potential to identify and exploit software vulnerabilities at unprecedented speed.