Anthropic's Mythos kept private

Anthropic is keeping its new Mythos model off the public market and giving it only to selected partners for security testing. (anthropic.com) The company announced Project Glasswing on April 7, 2026, with launch partners including Amazon Web Services, Apple, Cisco, CrowdStrike, Google, JPMorganChase, Microsoft, NVIDIA, Palo Alto Networks, Broadcom, the Linux Foundation, and Anthropic itself. Anthropic said it also extended access to more than 40 additional organizations that build or maintain critical software infrastructure. (anthropic.com) Anthropic said Mythos Preview is an unreleased general-purpose model that is unusually strong at computer security work, including finding and exploiting software flaws. In a technical post the same day, the company said the model identified zero-day vulnerabilities in every major operating system and every major web browser during testing. (anthropic.com) A zero-day is a bug the software maker does not know about yet, which means there is no patch when it is first found. Anthropic said more than 99% of the vulnerabilities it found remain unpatched, so it withheld most technical details under coordinated disclosure rules. (anthropic.com) Anthropic said Mythos Preview has already found thousands of high-severity vulnerabilities, and it is committing up to $100 million in usage credits plus $4 million in donations to open-source security groups through Glasswing. The stated plan is to let defenders scan first-party and open-source systems before similar capabilities spread more widely. (anthropic.com) The company’s public explanation is that coding systems have crossed a threshold where they can outperform nearly all human experts at spotting and exploiting some vulnerabilities. Anthropic wrote that the model can “surpass all but the most skilled humans” at that work, which is why it chose controlled access instead of a broad release. (anthropic.com) Outside reporting shows Anthropic’s caution is not universal industry consensus, even if many security researchers take the claims seriously. NBC News reported that some experts urged caution because Anthropic has not publicly documented most of the bugs, while Luta Security chief executive Katie Moussouris said the risks are “very much real.” (nbcnews.com) Anthropic is also using the restricted rollout to gather evidence about how the model behaves in real defensive work. On its Glasswing page, the company said partners will use Mythos Preview in their own security operations and Anthropic will share lessons from that work with the broader industry. (anthropic.com) That leaves Mythos in an unusual position for a frontier model in 2026: announced publicly, described as the company’s most capable system yet, but distributed like a sensitive security tool instead of a normal product launch. (anthropic.com)