Anthropic is moving to restore access to Claude Mythos 5, its frontier model with standout cybersecurity capabilities, for a limited set of entities approved by the US government. The model, which had been suspended under export controls, will reopen only to trusted partners.
June 27, 2026 · Anthropic
U.S. Reopens Anthropic's Most Powerful Cyber Model — to a Chosen Few
After government export controls froze all access on June 12-13, Anthropic can again deploy Claude Mythos 5 — but only to approved organizations defending U.S. critical infrastructure.
1,000s
zero-day vulnerabilities it can find & exploit across major OSes and browsers
100+
approved U.S. companies & agencies in the limited Project Glasswing program
Hours
to surface flaws in classified systems that evaded review for decades
How access swung
Jun 12-13
Export controls bar foreign access to Mythos 5 & Fable 5
→
Then
Anthropic suspends access for all customers
→
Jun 27
Restored — but only to approved infrastructure defenders
Same base model · two very different doors
Claude Mythos 5
SAFETY CLASSIFIERS REMOVED
Cyber-focused vulnerability finding & defense
Approved Glasswing orgs only
Not publicly released
Claude Fable 5
SAFETY MECHANISMS INTACT
General work & development
Broadly to enterprise & paid subscribers
Publicly released
A "GAME CHANGER" FOR DEFENSE
In government testing it found vulnerabilities in classified systems within hours, outperforming earlier models on CTF and multi-stage attack simulations.
DEFENDERS LEFT DISARMED?
Dozens of experts warn the controls create market uncertainty and undermine U.S. AI leadership. A jailbreak case is cited as one trigger for regulation.
A compromise between powerful cyber capability and the risk of misuse — general access to Fable 5 and any wider Mythos 5 rollout remain under negotiation.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…