Anthropic will restore access to Fable 5 after the lifting of US restrictions - ForkLog

# Anthropic will restore access to Fable 5 after US restrictions are lifted

The US Department of Commerce removed export restrictions on Claude Fable 5 and Claude Mythos 5. The Anthropic team reported this on June 30.

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, and will share an update soon.

We’re grateful to our users for their patience, and to everyone who worked with us on…

— Anthropic (@AnthropicAI) June 30, 2026

The company will open Fable 5 to users worldwide on July 1. The model will be available on Claude Platform, Claude.ai, Claude Code, and Claude Cowork. The ability to work through AWS, Google Cloud, and Microsoft Foundry will be restored «as quickly as possible,» Anthropic said.

Claude Fable 5 will be available again globally tomorrow.

After a series of productive conversations with the US government, we're redeploying the model with a new set of classifiers to target and block more cybersecurity tasks. In the near term, some routine tasks like coding…

— Anthropic (@AnthropicAI) July 1, 2026

*«After a series of productive conversations with the US government, we are resuming deployment of the model with a new set of classifiers to more precisely identify and block tasks related to cybersecurity. […] We have also begun developing a consensus framework — together with Amazon, Microsoft, Google and other Glasswing partners — to assess the severity of AI jailbreaks and determine how AI developers should respond to them,» — the post states.

Access to Mythos 5 has already been restored for some US organizations after government approval on June 26. The company will continue to cooperate with authorities. According to Anthropic representatives, this will include pre-release access to tools and protective measures for assessment, exchange of information about breaches and abuses, and dedicated resources for joint research.

To restore Fable 5, developers trained a new safety classifier: it should limit the behavior described in Amazon's report to the authorities. If a request is blocked, the user will receive a notification, and the request itself will be redirected to Claude Opus 4.8. According to the company, the new classifier detects the specific method in more than 99% of cases.

«It is probably impossible to make any AI model completely resilient to jailbreaks,» — Anthropic stated.

The company acknowledged that the new measures will lead to more false positives in normal programming and debugging tasks. The team called this a compromise for broader access to the model's other capabilities.

In addition, Anthropic, together with Amazon, Microsoft, Google and other partners of Project Glasswing, has begun developing a common approach to assessing the danger of jailbreaks. The company noted that currently there is no single standard in the AI industry for determining the severity of such security bypasses.

The proposed scheme evaluates a jailbreak according to four criteria:

• how much it expands the attacker's capabilities compared to available tools;
• how many different malicious tasks the bypass applies to;
• how easy it is to turn it into a real attack;
• how easy it is for other people to find or repeat this method.

For the most dangerous cases, the company promises to begin preliminary protective measures immediately after confirming the severity of the threat. Anthropic is also creating a team for round-the-clock monitoring of channels where data about jailbreaks is published, and is launching a HackerOne program for reporting possible bypasses of Fable 5's protection.

Recall that in June, amid restrictions on Anthropic, OpenAI launched the full version of a specialized model for finding, checking, and fixing vulnerabilities, GPT-5.5-Cyber.

At the end of the month, at the request of US authorities, the company opened limited access to GPT-5.6 Sol, Terra, and Luna for a small group of trusted partners.