Unauthorized Access Reported for Anthropic's Mythos Cybersecurity Tool
Anthropic is investigating reports that an unauthorized group has gained access to Mythos, its highly sensitive cybersecurity AI model. According to reports from Bloomberg and TechCrunch, members of a private online forum were able to access the tool through a third-party vendor shortly after its announcement. While Anthropic maintains that its internal systems remain secure, the group has reportedly provided screenshots and live demonstrations of the tool in use.
Mythos, part of Anthropic's 'Project Glasswing,' is a frontier model designed to identify and weaponize software vulnerabilities. It was intended for a strictly controlled research preview via Amazon Bedrock, limited to an initial allow-list of organizations. The reported breach raises significant concerns regarding the containment of powerful AI tools capable of automating cyberattacks.
Key Takeaways
- Unauthorized users reportedly gained access to Anthropic's Mythos cybersecurity tool on the same day it was announced.techcrunch.comanthropic.com
- The access was allegedly achieved by leveraging a third-party contractor's credentials and guessing the model's online location.anthropic.com
- Mythos is a highly sensitive 'frontier model' capable of identifying and weaponizing critical software vulnerabilities.nbcnews.comamazon.com
- Anthropic is investigating the reports but claims there is currently no evidence that its internal systems were compromised.techcrunch.com
- The tool was intended to be restricted to a select group of partners via Amazon Bedrock's gated research preview.bloomberg.com
Affected Companies
Sources
Unauthorized group has gained access to Anthropic's exclusive cyber tool Mythos, report claims
TechCrunch
“Anthropic is investigating claims of unauthorized access while maintaining that their internal systems show no signs of impact.”
Project Glasswing
Anthropic
“The Mythos model is designed to identify and weaponize vulnerabilities to help defenders secure critical software.”
Anthropic’s Project Glasswing Mythos Preview Claude gets limited release
NBC News
“Researchers warn that the Mythos model can detect and weaponize thousands of high-severity software bugs.”