Brief
Palo Alto Networks tests Anthropic's Mythos AI, finds major security flaws
A controlled test in April showed Anthropic's Mythos model could quickly expose weaknesses in Palo Alto Networks' infrastructure.
By Felo News Desk · Published
In April, Palo Alto Networks CEO Nikesh Arora allowed Anthropic's unreleased AI model Mythos to scan the company’s internal systems, according to Fortune. The test revealed numerous vulnerabilities that the model identified at a speed and scale the firm described as unprecedented.
Mythos, a frontier AI model for advanced coding and cybersecurity, had previously drawn U.S. government scrutiny, prompting temporary export controls in June that forced Anthropic to disable the model for all users. The government later restored limited access for vetted U.S. users and select users in 15 other countries.
Arora said the results underscored a new era for cybersecurity, where AI could both empower defenders and empower attackers to locate critical flaws almost instantly. He noted that CEOs are now demanding demonstrations of such AI capabilities, marking a shift from previous attitudes toward cyber risk.
Key facts
- Anthropic's Mythos AI was tested on Palo Alto Networks' internal infrastructure in April. (fortune.com)
- The test uncovered significant vulnerabilities at a speed and scale described as unprecedented. (fortune.com)
- U.S. export controls temporarily disabled Mythos in June before limited access was restored. (fortune.com)
Sources
- [1] fortune.com — originally reported as “Palo Alto Networks' Nikesh Arora is building a defense against the dark side of AI”
