Anthropic has been quietly preparing its most powerful model for a wider release, and the evidence is now stacked enough to write about. Last weekend, the strings "claude-mythos-1-preview" appeared briefly inside the live Claude interface before being pulled, spotted first by TestingCatalog and confirmed across multiple security outlets in the days since. The labels point to two specific access points, Claude Code and a redesigned Claude Security dashboard. On May 22, Anthropic itself posted on X that "in the near future, once we've developed the far stronger safeguards we need, we look forward to making Mythos-class models available through a general release." Read together, those two things are the company telegraphing what comes next: the model the rest of the world has not been allowed to use is being prepped for enterprise customers, and the dashboard around it is being built right now.

This is the next chapter of a story we have been tracking since March, and for your audience, it matters because Mythos is not a normal model launch. It is the model Anthropic itself called too dangerous to ship freely, and the loosening of that grip is the actual news.

A short recap to ground anyone catching up. Mythos surfaced on March 26 when a misconfigured Anthropic content management system exposed roughly 3,000 unpublished internal assets, including a draft blog post describing the model as "by far the most powerful AI model we've ever developed." Anthropic confirmed it days later, calling Mythos a "step change" in capabilities for cybersecurity, coding, and academic reasoning.

Instead of a general release, Anthropic launched Project Glasswing on April 7, a tightly controlled program giving access to roughly 40 partner organizations, including AWS, Microsoft, Google, Apple, NVIDIA, JPMorgan Chase, CrowdStrike, and the Linux Foundation. The reasoning was openly defensive: Mythos can find security vulnerabilities at superhuman speed, and Anthropic wanted to make sure defenders had it before attackers caught up.

The results justify the caution. Glasswing partners, plus Anthropic's own teams, have now found more than 10,000 high or critical-severity vulnerabilities in essential software. Anthropic's public coordinated disclosure dashboard, updated May 22, lists 1,596 specific disclosures across 281 open-source projects with 97 patched so far. This is not a chatbot. It is a vulnerability-finding engine that has already shifted the global security baseline.

That history is exactly why the Mythos 1 sighting matters. The model is leaving the controlled environment.

Here is what the public signals add up to.

The model ID: "claude-mythos-1-preview" briefly visible inside Claude. Not a developer build, not a private console, the live consumer interface. That is the kind of accidental exposure that happens when a release is being staged.

The new access points: Claude Code and Claude Security. Code is the developer terminal Anthropic has been turning into its agent platform. Security is the enterprise product where the new dashboard is being built, including seven-day and thirty-day historical charts, deeper triage, and a unified vulnerability view.

Anthropic's own words, on the record, May 22: "We look forward to making Mythos-class models available through a general release." That is the company explicitly committing to broader access once safeguards are ready.

The strategic pattern: Anthropic's Opus 4.7 rollout already included built-in cybersecurity safeguards that automatically detect and block high-risk offensive use. Multiple researchers and reporters have flagged that as the "stepping stone" toward a Mythos-class release. The architecture for safely shipping this is being built in pieces, in public.

So the honest read is not "Mythos launches tomorrow." It is "Anthropic is constructing the channel through which Mythos reaches paying customers, and it is going to be Claude Code and Claude Security first, not a free chat product."

Three reasons that matter for an audience that allocates capital and runs teams.

First, the model that scared a government is becoming a product. The UK NHS walled off GitHub repositories because of Mythos. The White House drafted an executive order around frontier model vetting partly because of Mythos. The Pentagon excluded Anthropic from classified networks partly because of Mythos. Now Anthropic is bringing that same capability to Claude Code subscribers and Claude Security customers, behind gates, with safeguards, but unmistakably wider than 40 partners. The economic logic is obvious. The model with the most dangerous capabilities is also the model enterprises will pay the most for.

Second, this is how dual-use AI gets commercialized in practice. Watch the pattern: leak forces transparency, restricted partner program proves defensive value, enterprise product captures the revenue, general release eventually follows under tightening safeguards. That is the playbook the entire industry is going to use for capabilities too sensitive for an open API. Mythos is the first real case. Anyone evaluating AI vendors should expect every frontier lab to copy this structure.

Third, the cybersecurity stack is being rewritten in real time. If Mythos 1 lands in Claude Security with the new dashboard, every enterprise security team has a choice to make. Continue running the existing vulnerability-management stack, or start integrating a model that has already found more critical bugs in major software than most security firms find in a year. The procurement conversation in serious CISOs' offices is about to change.

The Anthropic round we covered Monday officially closed this week. The deal exceeded $30 billion at a pre-money valuation above $900 billion, formally making Anthropic the world's most valuable private AI company, surpassing OpenAI's $852 billion March mark. The round was co-led by Sequoia, Dragoneer, Altimeter, and Greenoaks, with Microsoft, NVIDIA, Founders Fund, and General Catalyst also participating. Multiple sources expect this to be Anthropic's final private round before an October public offering.

Two things to take from the close. One, the valuation jump from $380 billion in February to over $900 billion in May is among the fastest at this scale in venture history, and the revenue numbers behind it ($11 billion projected Q2, first operating profit) actually support it. Two, the Mythos commercialization story above and the funding close are not separate. The investor pitch for a $900 billion valuation requires showing what Anthropic ships next that justifies it. Mythos 1 in Claude Code and Claude Security is that answer.

The string "claude-mythos-1-preview" briefly appeared inside Claude this past week, alongside new code references and a redesigned Claude Security dashboard, and Anthropic itself tweeted on May 22 that Mythos-class models are coming to a general release once safeguards are ready. The release path is becoming visible: Mythos 1 lands first in Claude Code and Claude Security for enterprise customers, not as a free public model. The capability behind it has already produced 10,000+ critical vulnerability findings and 1,596 coordinated disclosures across 281 open-source projects. Meanwhile Anthropic's $30 billion round at a $900 billion-plus valuation officially closed this week, making it formally the world's most valuable private AI company and setting up the October IPO. Both stories are the same story. The most powerful model ever built is about to start generating the revenue that justifies the most valuable AI company ever priced. The cage is opening, on Anthropic's terms.

Stay building. 🤖