Anthropic Names Its Mythos Public Version: Claude Fable Lands as Early as June 10
Anthropic is set to release the public version of Mythos as early as June 10, according to reporting by Alex Heath at Sources. The public model will arrive under a new name: Claude Fable. The launch follows months of restricted access through Project Glasswing, which limits Mythos to roughly 200 vetted organisations across critical infrastructure, operating systems, and enterprise software.
The rebrand signals substantive changes to the model’s guardrails, not merely its packaging. The public-facing version will carry “significantly stricter safeguards” than the Glasswing preview. A separate red-team model, claude-oceanus-v1-p, surfaced in inference infrastructure this week, indicating Anthropic’s safety pipeline is actively loaded.
The Exploit Research Behind the Delay
Anthropic’s rationale for the phased rollout rests on concrete capability measurements published alongside the launch window. Two evaluations stand out.
Firefox patches: Mythos Preview autonomously developed 8 working code-execution exploits from 18 Firefox security patches. The model produced its first exploit in under one hour of patch release. The full Firefox release containing the fix was still 18 days away at the time of testing, extending the exposure window substantially.
Windows kernel vulnerabilities: Against 21 Windows kernel vulnerabilities disclosed between January and February 2026, Mythos generated proof-of-concept exploits for 18. It produced its first PoC in 31 minutes and completed all 18 within six hours. Eight exploit chains ultimately achieved SYSTEM-level privilege escalation from a low-privileged user account. The Windows evaluation used closed-source software, requiring the model to work from patched binaries and decompiler output rather than source code.
Anthropic’s researchers put it directly: “N-day has become dangerously misleading. N-hour is closer to the reality we now operate in.”
Why Now
Anthropic stated at the Opus 4.8 launch that “Mythos-class models roll out to all customers in coming weeks,” reversing its April position that no general release was planned. The company has consistently framed the timeline around safeguard maturity rather than commercial pressure.
The structural argument for releasing: within six to twelve months, Anthropic expects multiple other labs to field Mythos-class models. A lab that waits indefinitely cedes the defensive capability to organisations with fewer safeguards. Fable is the instrument for getting the defensive side of that equation into production before that window closes.
Project Glasswing now spans more than 200 organisations across 15 countries, including Apple, Google, Microsoft, and major financial institutions. Anthropic estimates that a successful attack on any single Glasswing partner’s codebase could affect more than 100 million people. Fable extends that defensive layer to any Claude customer with appropriate verification.
What Changes at Public Launch
The Fable version retains Mythos-class intelligence but with reduced offensive ceiling and tighter usage controls. Glasswing partners retain priority access to broader capabilities through the Cyber Verification Program. Anthropic plans to scale that program to additional organisations specifically cleared for cyberdefense work.
The commercial implication is direct: Claude Fable is the first frontier-class model built specifically for the security market to reach general availability. Competitors including OpenAI’s GPT-5.5-Cyber remain gated to narrower distribution. If Fable launches tomorrow on schedule, the security tooling gap between Anthropic and the rest of the market widens further before it narrows.