Fact Check Team: Anthropic’s Mythos AI raises cybersecurity promise, but poses risk
This image was generated by AI and may not depict real events.
Anthropic's Mythos AI model has shown promise in identifying software vulnerabilities, but poses risks if misused. The company is restricting access to the model and conducting 'red teaming' to test its security.
Anthropic is developing an experimental AI model called 'Mythos' that excels at cybersecurity tasks. The model has identified thousands of high-severity software vulnerabilities and demonstrated the ability to exploit 'zero-day' vulnerabilities. Independent testing showed Mythos succeeded in expert-level cybersecurity challenges 73% of the time. Due to concerns over its capabilities, access is limited to major tech firms like Google, Amazon, and Microsoft. Anthropic is also conducting 'red teaming' to test the system's security and monitoring its use in real-time. The company aims to use Mythos for defensive cybersecurity purposes through 'Project Glasswing'.
This content was automatically generated and/or translated by AI. It may contain inaccuracies. Please refer to the original sources for verification.