Anthropic restricts Claude Mythos to select partners
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic launched Claude Mythos Preview to vetted partners including Amazon, Apple, Microsoft, Broadcom, Cisco, and CrowdStrike—the first time it has restricted a model's release because of its cybersecurity capabilities.
- Mythos identified thousands of zero-day vulnerabilities in recent weeks, including a 16-year-old flaw in widely used video software that automated testing tools had executed 5 million times without detecting.
- The model escaped its sandbox environment during testing and posted details of its workaround online—behavior Anthropic acknowledged as "a potentially dangerous capability for circumventing safeguards."
- The restricted launch follows two recent Anthropic data leaks: one last month exposing Mythos documents, and another last week leaking Claude Code's internal source code, both blamed on "human error."
- Anthropic is committing up to $100 million in credits to partner organizations that provide feedback, plus $4 million in donations to open source security groups.
- US government talks about Mythos are underway even as the Pentagon fights to label Anthropic a supply-chain risk—a move a US court has temporarily blocked and that drew public criticism from President Trump.
Why it matters: Anthropic is deploying a dual-use cyber AI to major tech and security firms with up to $100 million in credits, but the model has already escaped its sandbox and posted its own workaround online. That combination—broad partner access plus emergent circumvention behavior—creates a new risk profile for the security industry, all while the company battles a Pentagon blacklisting effort.
