Mythos 1 Heading Towards Release: Traces of the Model in Claude Code
In recent days, some users of Claude Code noticed the appearance and then disappearance of a model option called Mythos 1, internally identified as claude-mythos-1-preview. The trace, captured in a screenshot published by TestingCatalog on May 23, indicates that Anthropic is preparing the commercial release of the model, which it has kept off the market so far for safety reasons.
Strings in the Code of Claude Code and Claude Security
The appearance of the item in the interface lasted only a few minutes, but the strings left in the source code are more explicit. One, in particular, attracts attention: "Access to the Claude Mythos model in Claude Code and Claude Security". Additional references to the model have appeared in the platform's infrastructure. Anthropic has not publicly confirmed either the integration or a release date.
ANTHROPIC 🔥: Mythos 1, "claude-mythos-1-preview", is being prepared for a release on Claude Code and Claude Security. The model became visible for a short amount of time on Claude; besides that, new strings mentioning Mythos have been added.
🚨 AI News | TestingCatalog (@testingcatalog) May 23, 2026
The two products mentioned are those that the company has already pushed towards the enterprise segment. Claude Code is the agentic environment for developers; Claude Security is the platform for scanning vulnerabilities in enterprise codebases, which recently entered public beta for Enterprise clients. Anthropic has stated that, powered by the current flagship model Claude Opus 4.7, the tool has helped fix over 2,100 vulnerabilities in three weeks. It is currently unclear whether Mythos 1 will be included in both products or confined to the Enterprise branch.
The Change in Tone in the Glasswing Report
The leak fits into a reversal of position communicated by Anthropic itself. The first official update of Project Glasswing, published on May 22, closes with a passage that modifies the company's position regarding the April announcement: the company states it is looking towards a near future in which to make the Mythos class models available to the public, once adequate safety barriers are developed. The model, described until a few weeks ago as too capable and "dangerous" to be put in the hands of the public, is now indicated as a possible candidate for general release.
The same update, which we discussed a few days ago, reports that the approximately 50 partners of the Glasswing consortium have identified over 10,000 high or critical severity vulnerabilities in a month of operation. This raises another issue, namely the ability to assess, disclose, and apply fixes to a volume of findings that grows more rapidly than human teams can manage.
The interesting point is the apparent contradiction between the caution declared at the presentation of the model and the trace that appeared in production. In the same document, Anthropic admits that no company, including itself, has yet developed sufficiently robust protections to prevent hostile use of models of this class. The presence of a preview identifier in the production strings suggests that the necessary safety barriers are in advanced development but not necessarily complete. The preview label, in Anthropic's customs, has always preceded the general release by several months.
Anthropic has not indicated a release date nor clarified if Mythos will be available on the public subscription plans of Claude or remain confined to enterprise products. The top model currently available to the public remains Claude Opus 4.7, with Sonnet 4.6 and Haiku 4.5 completing the lineup.