Anthropic's most capable AI model gets a controlled rollout to roughly 200 vetted groups, with national security concerns already shaping what comes next
Anthropic announced Claude Mythos Preview on April 7, 2026, under a program called Project Glasswing. The initial cohort was capped at approximately 200 organizations, including US government entities.
What Mythos actually does, and why that matters #
Mythos models demonstrated the ability to find thousands of vulnerabilities across major operating systems and browsers, including zero-day vulnerabilities. Anthropic paired the launch with a financial commitment of up to $100 million in usage credits for participating organizations, plus $4 million in direct donations to support the initiative.
Collaboration with AWS through Amazon Bedrock was part of the gated preview from day one, with Microsoft and Google also involved in the broader technology partnership framework.
On June 9, 2026, Anthropic released Claude Mythos 5 to select partner organizations while simultaneously making Claude Fable 5 available to the general public. The June 9 release followed a voluntary 30-day federal review period, a process aligned with policies stemming from the Trump administration’s executive order on frontier AI models.
The security scare that nobody wanted #
A mid-June 2026 emergency directive from the US government temporarily suspended access to the new AI models, citing national security concerns tied to a reported jailbreak vulnerability.
What this means for investors and the broader AI market #
The jailbreak incident adds a wrinkle to valuations across the AI sector. If a model as carefully deployed as Mythos, with a vetted rollout, federal review, and major cloud partnerships, can still trigger an emergency suspension over a security concern, the implied risk for less carefully managed deployments is considerably higher.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our