Breaking

US Lets Anthropic Reopen Mythos 5 to Trusted Organizations

June 26, 2026 at 22:33 EDT

Policy & Regulation
Foundation Models
Security

On June 26-27, 2026, the US government allowed Anthropic to redeploy its most advanced AI model, Claude Mythos 5—suspended over national security concerns—to a set of "trusted" US organizations that operate and defend critical infrastructure, partially restoring access. The move was confirmed by Anthropic's official statement and Reuters reporting.

June 26, 2026 · Anthropic × US Commerce Dept.

US Partially Lifts Export Curbs on Anthropic's "Mythos 5" Cyber Model

Commerce Secretary Howard Lutnick reversed part of a June 12 suspension, clearing the frontier cybersecurity model for redeployment to 100+ "trusted" US organizations defending critical infrastructure — after talks concluded "appropriate safeguards are in place."

100+

Trusted US orgs cleared (Annex A)

~14

Days suspended
(Jun 12 → Jun 26)

~90%

True positives of 1,900 detections (early field test)

Timeline: launch → suspension → partial release

A frontier AI treated as "weapons-grade" under US export-control policy.

JUN 9

Mythos 5 unveiled

Cyber-focused sibling of Claude Fable 5, via Project Glasswing.

→

~JUN 12

Full suspension

Export order halts foreign access; both models disabled.

→

JUN 26

Partial release

License-free redeploy to trusted infrastructure defenders.

Mythos 5 benchmark scores

95.5%

SWE-bench
Verified

80.3%

SWE-bench
Pro

88.0%

Terminal-
Bench 2.1

94.1%

GPQA
Diamond

78.5%

ArXiv
Math

Supporters

Praise the dual structure — a public, guarded Fable 5 plus a restricted, full-capability Mythos 5 that lets trusted defenders find OS/browser zero-days and CVE exploits fast and cheap.

Critics

Warn that abrupt government suspension and release undermine predictability in AI development, raise misuse risk, and harm foreign employees cut off from access.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.