AI has the power to make or break cybersecurity. Aiming to prevent the worst-case scenarios that their ever-more-powerful models could cause, providers are scrambling to find solutions.

On Thursday, Anthropic released Claude Security, a dedicated offering for weeding out code vulnerabilities, in public beta for Enterprise users. Though not as powerful as Claude Mythos, which Anthropic said can surpass “elite human experts” at finding holes in software, the product is powered by Opus 4.7, the company’s most powerful generally available model, and can be leveraged by a much wider set of organizations.

With AI models becoming more capable of discovering vulnerabilities, it’s only a matter of time before their capabilities are put to use to exploit them, too, Anthropic said in its post. “Now is the time for organizations to act to improve their security, preparing for a world in which working software exploits are much easier to discover. ”

The product, previously Claude Code Security, was released in a limited research preview in February and has since been tested by hundreds of organizations, the company reported.

Rather than searching for known patterns, Claude Security reasons about code the same way a security researcher would, Anthropic says.

• Claude Security offers scheduled and targeted scans of codebases, integration with audit systems, and the ability to track triaged findings.
• The tool also doesn’t require API integrations or custom agent buildouts. So long as an enterprise uses Claude, they can leverage Claude Security.
• Anthropic is also partnering to integrate Claude Security into several existing software tools, including CrowdStrike, Microsoft Security, Palo Alto Networks, SentinelOne, TrendAI, and Wiz.

Claude Security is being released as the White House seeks to leash Anthropic from widening the reach of Claude Mythos, its most powerful model yet. On Thursday, the Wall Street Journal reported that the administration opposes Anthropic’s plan to release Mythos to around 70 more organizations, bringing the total number to 120 entities. Sources told WSJ that the opposition is largely due to security concerns.

However, OpenAI may be breathing down Anthropic’s neck. Research released on Thursday by the AI Security Institute found OpenAI's GPT-5.5 cybersecurity capabilities are comparable to Mythos Preview, claiming the model is “one of the strongest models [they] have tested” based on its cyber capabilities.

Our Deeper View

Anthropic and OpenAI both seem to be playing with fire. Anthropic is muzzling Mythos because it is a seemingly double-edged sword, capable of both supporting and completely upending cybersecurity and even national security. Offering Claude Security on Opus 4.7 is a bit of a consolation prize for those who aren’t the elite few given access to Mythos. OpenAI, it seems, is taking a similar path, with CEO Sam Altman announcing on Thursday that GPT-5.5-Cyber, its frontier security model, will be available only to “critical cyber defenders” for now. At the end of the day, both companies are aiming to get ahead of problems that their models are effectively poised to worsen. Whether their solutions will be stitches or Band-Aids is uncertain until these powerful models are out in the world. And that's the scary part.