"The advantage will belong to the side that can get the most out of these tools," Anthropic warned in April when it announced the Mythos model.
Anthropic's April warning and the decision to restrict Mythos
In April, Anthropic announced the Mythos model as a restricted offering and made it available only to a limited set of partners, including security researchers. At the time the company explicitly cited major "security" risks and decided against a broad public rollout. That caution was framed around a short-term risk that attackers could gain advantage if frontier labs released very capable models without adequate protections, and a longer-term expectation that defenders would eventually use such tools to harden systems: "In the short term, this could be attackers, if frontier labs aren’t careful about how they release these models. In the long term, we expect it will be defenders who will more efficiently direct resources and use these models to fix bugs before new code ever ships."
Why Anthropic delayed the public release
Anthropic's temporary gating of Mythos followed a common pattern among AI developers: companies typically hold back powerful models until they build "strong guardrails" to reduce the potential for misuse. The company framed the April pause as a security-driven choice, prioritizing risk mitigation over immediate public availability.
Anthropic now signals a public rollout is imminent
In a new blog post, Anthropic confirmed it plans to bring Mythos-class models to the general public in the coming weeks, though it did not commit to a specific calendar date. "We’re making swift progress on developing these safeguards and expect to be able to bring Mythos-class models to all our customers in the coming weeks," Anthropic wrote. The company also said it is already permitting a small number of organizations to use a Claude Mythos preview for cybersecurity work, but the company did not confirm whether the preview instance is the same model that will be released more broadly.
Claims about Mythos capabilities and the Claude Code preview
Anthropic says the Mythos model shows major improvements in code reasoning and autonomy, and the company or the reporting characterizes it as being believed to be far more powerful than Opus 4.8 and other models available on the internet. Separately, a "Mythos-preview" model briefly appeared for some users on Claude Code before it was taken offline, a detail the company and reporting both recorded.
What this means for technologists, enterprises, and adversaries
- Technologists and security teams: Anthropic has already allowed a small number of organizations to use the Claude Mythos preview for cybersecurity work, indicating early operational testing by defenders. If the company's claim of improved code reasoning and autonomy holds, those teams may gain new tools for vulnerability discovery and remediation.
- Procurement and affected enterprises: Anthropic's statement that it expects to bring Mythos-class models to "all our customers in the coming weeks" signals a pending availability decision buyers will need to track closely, although no firm timeframe was given and it remains unclear whether the preview matches the public version.
- Adversaries and threat actors: The company itself warned that, in the short term, attackers could gain advantage if frontier labs release powerful models without adequate safeguards — a risk Anthropic said factored into its earlier decision to restrict Mythos.
Anthropic's move from restriction to an imminent public rollout marks a clear transition from caution toward broader distribution, but the company has left two practical questions open in its public statements: whether the preview used by a small number of organizations is the same model slated for public release, and exactly when that release will occur. Anthropic says it is making "swift progress" on safeguards and expects public availability in the coming weeks; until the company publishes a specific timeframe or a definitive model specification, observers will have to assess the rollout on the company's subsequent disclosures and any differences between the preview and the publicly released Mythos-class models.
