Skip to content

Anthropic

Held-back Mythos 5 now scans enterprise code

3 min read

Written by an AI model from the linked sources and checked against them by a second model. How we work

Image: horizonsmedia.fr

Anthropic on August 21 opened Claude Mythos 5, the model it had kept out of public release, to every Claude Enterprise customer, The New Stack reported. There’s a catch: the only way in is Claude Security, the code-scanning tool Anthropic launched earlier this year.

Before this, Mythos 5 was limited to about 150 partners in Project Glasswing. Anthropic set up Glasswing on April 7 with Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA and Palo Alto Networks, after seeing what the unreleased Mythos Preview could do. On June 2 it added 150 partners in more than 15 countries, across power, water, healthcare, communications and hardware, and each had to meet security requirements first, per CNBC. Everyone else got Fable 5 in June. Fable 5 is essentially Mythos 5 with very strict guardrails, and the U.S. government banned it and then lifted the ban.

In Claude Security, Mythos 5 runs as a public beta. An admin turns it on, and then developers and security teams scan their repositories with Mythos 5 doing the work and suggesting fixes for whatever it finds. Nobody gets to prompt the model directly. Users only see what it produces, and Anthropic gives its reasoning in the announcement:

“The riskiest behavior occurs when a user has direct access to a model, where a malicious actor can try to steer it toward harmful uses. But if users can only receive specific outputs, such as a patch for a vulnerability or a security alert, that risk is much lower.”

The company says the changes give users “greater access to the defensive results, while maintaining appropriate guardrails around direct access to the model.” Everything runs inside Anthropic’s own tools and harnesses, and Claude Security “uses Mythos 5 to scan code you own.” Anthropic adds that it and its partners “have abuse prevention measures in place to verify the model stays within its intended scope.”

The New Stack points out that ownership gets slippery once open source is involved, because so many codebases pull in open source libraries. A developer who clones a widely deployed library into a company repo and runs Mythos 5 on it would get exactly the findings an attacker wants, and that bug exists everywhere the library ships. So a lot of the safeguard rests on the phrase “code you own”.

The scanner bills at Mythos 5 rates: $10 per million input tokens and $50 per million output tokens.

Claude Security isn’t the only route Anthropic is opening. It is also helping other cybersecurity companies build Mythos 5 into their own products. Then there’s the Cyber Verification Program, which lets vetted defenders run dual-use security work on Opus and Sonnet with fewer blocks. Anthropic says those organizations will also get safeguarded access to Claude Mythos soon.

Open source gets its own money. The new Defender Advantage Fund (0xDAF) puts $35 million in credits toward finding and patching vulnerabilities in open source software. It follows the commitments Anthropic made when Glasswing launched in April: up to $100 million in Mythos Preview usage credits and $4 million in direct donations to open source security organizations.

No comments yet. Be the first.

Join the conversation

Sign in to leave a comment. It appears after moderation.