Anthropic has made its Mythos AI model publicly available, but with a catch: the company has attached explicit restrictions on cybersecurity-related applications. The release, confirmed by the company, means developers and researchers can now access the model's capabilities as long as they agree to limits on how it can be used in security contexts.
What the restrictions cover
The restrictions are specifically aimed at cybersecurity uses. The company did not detail every prohibited activity, but the move signals a push to prevent the model from being deployed for offensive security work, such as automated vulnerability scanning, penetration testing, or generating exploit code. Users will need to comply with these terms to access the model through Anthropic's platform.
Why the limits matter
Mythos is a large language model, and its public release opens the door for a wider range of experimentation. By placing explicit limits on cybersecurity, Anthropic is trying to balance openness with safety. The approach is not new for the company, which has previously focused on responsible AI deployment. But the public nature of this release means more eyes on how the restrictions are enforced.
Who gets affected
The restrictions will primarily affect developers and security researchers who might otherwise use the model for tasks like red-teaming or ethical hacking. Those with legitimate need to apply Mythos in cybersecurity roles will have to work within Anthropic's usage policy. The company hasn't said whether exceptions or special licenses are available.
The Mythos model's capabilities in other domains remain unrestricted, so the release still offers a powerful tool for general language tasks, content generation, and code writing — just not in cybersecurity.
Anthropic has not announced any further updates to the restrictions or a timeline for when they might change. For now, anyone wanting to use Mythos will have to read the fine print on cybersecurity.
