**Why did the US government restrict Anthropic's Fable 5 and Mythos 5 models?**

The government cut off access to both models on June 12 after discovering vulnerabilities in the safeguards designed to prevent misuse. The Trump administration cited national security concerns, and some researchers feared the models could be exploited to bypass cybersecurity measures. The restrictions remained in place for roughly three weeks before being lifted.

**When were Fable 5 and Mythos 5 restrictions lifted?**

US organizations had Mythos 5 access restored on June 26. Fable 5 became available globally shortly after Anthropic announced it had received notice from the Department of Commerce lifting the export controls. Anthropic posted on X that it would begin restoring access the following day.

**What did Anthropic agree to in order to get the export controls lifted?**

Anthropic agreed to expand its partnership with the US government. It also set up a red-teaming program with hackers to test its models and created a dedicated internal team to monitor emerging jailbreak threats around the clock. Commerce Secretary Howard Lutnick acknowledged Anthropic had taken steps in coordination with the government.

**What is the Glasswing program at Anthropic?**

Glasswing is an Anthropic program that gives cybersecurity researchers at trusted companies access to Mythos 5 for defensive purposes. After the export controls were lifted, Anthropic said it is working with the government to expand Mythos 5 access to a broader set of domestic and international partners through the Glasswing program.

**What is GRAM and what does it do for AI safety?**

GRAM, or Gradient-Routed Auxiliary Modules, is a pretraining method developed by AE Studio in collaboration with Anthropic. It isolates dangerous knowledge to specific modules in a language model that can be switched on or off. Experiments ran on models from 50 million to 5 billion parameters. Anthropic noted the research is preliminary and has not been applied to production models.

Anthropic Fable 5 & Mythos 5 Export Ban Lifted

By iCharles News · July 12, 2026

Bookmark

The Commerce Department lifted export restrictions on Anthropic's two most powerful AI models, Fable 5 and Mythos 5, about three weeks after flagging them as national security risks.

Anthropic Fable 5 & Mythos 5 Export Ban Lifted

What happened with Anthropic's Fable 5 and Mythos 5 export controls?

The US Department of Commerce lifted export restrictions on Anthropic's Claude Fable 5 and Mythos 5 models, roughly three weeks after flagging them as national security risks. Anthropic confirmed the news in a post on X: "We've received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5. We'll begin restoring access tomorrow."

Commerce Secretary Howard Lutnick told Anthropic in a letter that the company would "no longer need a license for exports or in-country transfers of its Claude Mythos and Claude Fable AI models," according to Ars Technica.

Why did the US government restrict the models in the first place?

The government abruptly cut off access to both models on June 12. The reason: officials discovered vulnerabilities in the safeguards designed to prevent misuse, TechXplore reported. Researchers feared the models could be exploited to bypass cybersecurity measures.

The Trump administration had invoked national security concerns to limit the release of advanced models from major US AI companies, including Anthropic.

When was access restored, and to whom?

Here's what we know so far about the timeline of restrictions and restorations:

RoguePlanet CVE-2026-50656: Defender Zero-Day

OpenAI Releases GPT-5.6 Sol & National Security

Prince George's County Passes 2-Year Data Center

GPT-5.6 Sol Restricted by U.S. Gov After

Date	Event
June 12	US government cuts off access to Fable 5 and Mythos 5
June 22	Small group of US cybersecurity firms authorized to access Mythos 5
June 26	US organizations have Mythos 5 access restored
July (announced)	Fable 5 available globally; Mythos 5 access expansion begins

Fable 5 is now available globally. Mythos 5 access was restored to US organizations on June 26, and Anthropic said it is working with the government to expand Mythos access to a "broader set of domestic and international partners."

What did Anthropic agree to in exchange for the lift?

Lutnick's letter acknowledged that Anthropic had "taken steps in close coordination with the US government to address the risks" posed by the models. Facing a longer delay, Anthropic agreed to expand its partnership with the government. The company also:

Set up a program to work with hackers to red-team its models
Created a dedicated internal team to monitor reports of emerging jailbreak threats around the clock

Lutnick reminded Anthropic in the letter that the US "reserves the right to re-evaluate the decisions" and reimpose export controls.

What is the Glasswing program?

Glasswing is an Anthropic program that allows cybersecurity researchers at trusted companies to access Mythos 5 for defensive purposes. Anthropic said it is now working with the government to expand Mythos access to a broader set of domestic and international partners through this program.

What is GRAM, and how does it relate to Anthropic's safety work?

Separately, researchers at AE Studio — in collaboration with Anthropic — published a method called Gradient-Routed Auxiliary Modules (GRAM), a pretraining technique that isolates dangerous knowledge to specific modules within a language model. Those modules can be switched on or off to control what a model knows, making it possible to restrict access to sensitive capabilities based on user trust level.

The research, posted to Anthropic's alignment blog, covers experiments on models ranging from 50 million to 5 billion parameters. The authors found that a single GRAM-trained model can approximate multiple models, each trained with a different category of dangerous data filtered out.

The four dual-use domains tested were:

Virology
Cybersecurity
Nuclear physics
Specialized code

Anthropic noted this research is preliminary and has not been applied to production models. The work was led by Ethan Roland, Murat Cubuktepe, and Erick Martinez at AE Studio, with contributions from Cem Anil and Alex Cloud at Anthropic.

This kind of AI safety research sits alongside the behavioral and access-control measures Anthropic already uses — including refusal training, classifiers, and tiered deployment — each of which the paper notes can be jailbroken or requires retuning for every release.

The broader context is a US government increasingly focused on AI national security risks, with companies like Anthropic now operating under agreements that give Commerce the right to reimpose restrictions at any time.

Anthropic's confirmed next step is expanding Mythos 5 access to a broader set of domestic and international partners through the Glasswing program.

Frequently asked questions

**Why did the US government restrict Anthropic's Fable 5 and Mythos 5 models?**: The government cut off access to both models on June 12 after discovering vulnerabilities in the safeguards designed to prevent misuse. The Trump administration cited national security concerns, and some researchers feared the models could be exploited to bypass cybersecurity measures. The restrictions remained in place for roughly three weeks before being lifted.
**When were Fable 5 and Mythos 5 restrictions lifted?**: US organizations had Mythos 5 access restored on June 26. Fable 5 became available globally shortly after Anthropic announced it had received notice from the Department of Commerce lifting the export controls. Anthropic posted on X that it would begin restoring access the following day.
**What did Anthropic agree to in order to get the export controls lifted?**: Anthropic agreed to expand its partnership with the US government. It also set up a red-teaming program with hackers to test its models and created a dedicated internal team to monitor emerging jailbreak threats around the clock. Commerce Secretary Howard Lutnick acknowledged Anthropic had taken steps in coordination with the government.
**What is the Glasswing program at Anthropic?**: Glasswing is an Anthropic program that gives cybersecurity researchers at trusted companies access to Mythos 5 for defensive purposes. After the export controls were lifted, Anthropic said it is working with the government to expand Mythos 5 access to a broader set of domestic and international partners through the Glasswing program.
**What is GRAM and what does it do for AI safety?**: GRAM, or Gradient-Routed Auxiliary Modules, is a pretraining method developed by AE Studio in collaboration with Anthropic. It isolates dangerous knowledge to specific modules in a language model that can be switched on or off. Experiments ran on models from 50 million to 5 billion parameters. Anthropic noted the research is preliminary and has not been applied to production models.

Verified claims

Each key claim below was checked against its source — the exact supporting passage is quoted so you can confirm it yourself.

Commerce Secretary Howard Lutnick told Anthropic the company would no longer need a license for exports or in-country transfers of its Claude Mythos and Claude Fable AI models.

no longer need a license for exports or in-country transfers of its Claude Mythos and Claude Fable AI models
Verified arstechnica.com
GRAM is a pretraining technique that isolates dangerous knowledge to specific modules within a language model that can be switched on or off.

Gradient-Routed Auxiliary Modules (GRAM)
Verified alignment.anthropic.com

Sources

according to Ars Technica arstechnica.com
TechXplore reported techxplore.com
Anthropic's alignment blog alignment.anthropic.com

Keep reading

GPT-5.6 Sol Is 54% More Token Efficient: Altman

Meta Muse Image Pulled After Privacy Backlash

OpenAI Loses No. 2 Exec & Head of Safety

Meta Muse Image: Opt Out Before It Uses Your