Anthropic's Mythos AI goes public June 9 with deliberate restrictions on vulnerability-finding capabilities, following a controlled rollout via Project Glasswing.
Anthropic is releasing a public version of Mythos, its highest-profile artificial intelligence model, today, June 9. The rollout follows weeks of restricted access through Project Glasswing, a selective program that limited the model to vetted banks and tech firms while Anthropic evaluated its security risks.
Journalist Alex Heath was first to report the release on X, writing that Anthropic was planning the public launch. A second source, going by Coined Bureau, claimed the product would ship under the name "Claude Fable." Anthropic has not confirmed either detail publicly.
Behind the timing sits a specific security concern. Mythos drew attention for its reported ability to identify critical vulnerabilities in web browsers and software systems. Executives at major financial institutions and officials at the IMF raised alarms, which pushed Anthropic to restrict access before any broader rollout.
Guardrails, not a lobotomy
The public Mythos will carry what Heath described as "substantial guardrails" and will not be "as cyber permissive" as the version Glasswing partners can access, according to The News. That language implies a meaningful reduction in offensive security capabilities without specifying where the cutoff sits.
On other dimensions, the release looks like an upgrade. Heath said the public version will be "dramatically better at long-run horizon and multi-run tasks" than prior Claude models, pointing to stronger agentic performance rather than unlocked security features.
Prediction markets have priced this as a near-certainty. Polymarket shows a 98% probability of Mythos releasing by June 30, and a 92% chance of a launch before June 15, per The News.
The awkward timing
Anthropic published statements earlier this week urging caution about AI self-improvement, warning that models capable of modifying their own code could outpace safety research. Releasing Mythos into that same news cycle creates friction the company will need to address directly.
Competitive pressure makes waiting difficult. Google is expected to open Gemini 3.5 Pro to the public this month, a frontier model with a 2-million-token context window and an integrated reasoning engine, as Her Zindagi reports. Both labs face the same calculus: ship fast enough to stay relevant, carefully enough to avoid a regulatory incident.
If Anthropic's tiered-access model proves durable, with full capabilities for credentialed institutions and a restricted version for general users, it would set the clearest precedent yet for graduated artificial intelligence deployment at scale, according to The News. The artificial intelligence review community and independent security researchers will test those guardrails quickly. How Anthropic responds to those findings, in public and on the record, will define the model's reputation more than today's launch.
Mythos ships into a landscape where trust in safety claims is low and scrutiny is immediate. The next two weeks will reveal whether Project Glasswing was a genuine risk-management framework or a delay tactic dressed in good intentions.
FAQ
What is Anthropic Mythos AI?
Mythos is Anthropic's latest model in the Claude family. It attracted scrutiny before its public launch for reportedly being able to identify critical vulnerabilities in web browsers and software systems at an unusual scale.
What is Project Glasswing?
Project Glasswing is Anthropic's controlled-access program that gave vetted financial institutions and tech companies early access to Mythos with fewer restrictions than the public version carries.
What is Claude Fable?
"Claude Fable" is the product name an X source attributed to the public Mythos release. Anthropic has not confirmed the name officially.
How does Mythos compare to Google Gemini 3.5 Pro?
Both are frontier models expected to reach the public in June 2026. Gemini 3.5 Pro emphasizes a 2-million-token context window and a "Deep Think" reasoning engine, per Her Zindagi. Mythos is positioned around long-horizon task performance and, in its restricted Glasswing form, advanced cybersecurity analysis.
