White House plans US agency access to Anthropic Mythos AI

The White House is preparing safeguards that could allow major US federal agencies to use Anthropic’s unreleased AI model Mythos for defensive cybersecurity work.

The move comes amid concerns that the model could also make it easier to find and exploit software flaws.

Mythos was announced on April 7 under Anthropic’s Project Glasswing, a limited program that lets select organizations use the Claude Mythos Preview model for defensive cybersecurity purposes.

In an email to Cabinet officials, federal CIO Gregory Barbaccia said the Office of Management and Budget was working with AI providers, industry partners, and the intelligence community on guardrails before any modified version is released.

A White House official said the model is still being evaluated for fidelity and security.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

Mythos could reshape defensive cybersecurity, but Anthropic is restricting access

Claude Mythos Preview is pitched as a way to find and fix software vulnerabilities at scale. Some third-party commentary says it could also create exploits on its own, while Anthropic frames partner access as focused on defensive security work 1.
Anthropic says Mythos Preview has uncovered thousands of previously unknown “zero-day” vulnerabilities, security flaws that were not known to software makers or defenders. This includes a 27-year-old bug in OpenBSD, an open-source operating system 2.
Anthropic is not offering Mythos Preview to the public because of dual-use risks. Access runs through Project Glasswing for vetted participants, including Apple, Google, and Microsoft 1.

Early access and policy tensions are shaping the rollout

Fortune reported that after its reporting on Mythos, shares in CrowdStrike, Palo Alto Networks, and other cybersecurity firms fell between 5% and 11% 2.
Third-party industry commentary argues that Project Glasswing members gain an early lead, while other businesses may take on more asymmetric risk 3.
Federal adoption described in the news is happening during reported contract and legal friction between Anthropic and the Pentagon, the US Department of Defense, according to a separate report 4.
An industry account says Anthropic is in a legal dispute tied to Pentagon demands for limits and guardrails on military uses of Claude, including autonomous weapons and mass surveillance 5.