US government suspends access to Claude Fable 5 and Claude Mythos 5 – AI News – #2 June 2026

The government’s executive action caught Anthropic completely off guard. Citing national security authorities, the directive was delivered to the company on Friday at 5:21 PM Eastern Time (ET). The immediate legal consequence of this order was the abrupt removal of the Fable 5 and Mythos 5 models from all consumer and commercial platforms.

It is important to emphasize that this suspension applies exclusively to this newly released, frontier tier of artificial intelligence. Access to older, highly capable models like Claude Opus 4.8 and Claude Sonnet remains entirely unaffected, allowing users to continue their workflows without disruption.

Official reports indicate that the US government became aware of a specific technique used to bypass the model’s safety filters (a “jailbreak”). According to officials, this method allowed the AI to identify software vulnerabilities, raising concerns that it could be weaponized by malicious actors.

After reviewing the demonstration used as the basis for the government’s order, Anthropic issued a formal statement challenging the decision:

From the initial launch, Anthropic openly acknowledged that deploying models with such advanced autonomous capabilities carries inherent risks. The company maintains that perfect, flawless resistance to jailbreaks is mathematically and technically impossible across the industry today. Because of this reality, Anthropic designed Fable 5 with a “defense in depth” philosophy.

Prior to its release, Fable 5 underwent thousands of hours of rigorous red-teaming evaluations in collaboration with the US government, the UK Artificial Intelligence Safety Institute (UK AISI), and independent third-party experts. These tests proved that Fable 5 possessed the most robust safety architecture ever deployed in a commercial model. Anthropic warns that if regulatory bodies recall commercial models over minor, non-universal vulnerabilities, it will effectively halt new model deployments for all frontier AI providers across the industry.

The sudden intervention by the US government has caused massive ripples in the tech sector precisely because Fable 5 and Mythos 5 offered capabilities that were previously considered years away. These models demonstrated an unprecedented ability to act as autonomous intellectual partners over long periods.

In pre-release trials, Fable 5 compressed months of complex engineering work for the payment platform Stripe into a single day. Operating autonomously, the model executed a system-wide migration across a massive Ruby codebase consisting of 50 million lines of code—a task that would have taken a human engineering team more than two months by hand.

The model also dominated complex analytical benchmarks, setting historical records in Hebbia’s senior-level financial reasoning tests. In computer vision, it proved capable of rebuilding a web application’s source code straight from screenshots and even successfully completed the classic game Pokémon FireRed by analyzing raw gameplay visuals alone, requiring none of the helper tools or navigation maps needed by older architectures.

Meanwhile, the specialized Claude Mythos 5 variant was already proving its worth in life sciences laboratories, accelerating the drug design process by roughly ten times. Working independently, the model successfully executed tasks traditionally handled by human scientists—such as choosing binding sites and running bioinformatics tools—yielding highly viable drug candidates for neurodegenerative and muscle diseases.

In genomics, Mythos 5 conducted a week of autonomous research, compiling single-cell data across 138 animal species to train a custom machine learning model. This AI-built system ultimately outperformed a recently published model in the prestigious journal Science, despite being 100 times smaller.

For subscribers of the Pro, Max, and Team plans, who were scheduled to enjoy complimentary access to Fable 5 until June 22, 2026, this directive means an immediate, forced rollback to Claude Opus 4.8.

Anthropic has issued an official apology to all customers and API developers facing severe disruptions to their operational pipelines. The company views the current recall as a profound misunderstanding resulting from a regulatory action that lacks transparency and technical grounding. Intensive discussions with US officials are currently underway to resolve the dispute, clarify the technical facts, and restore global access to the Mythos-class models as quickly as possible.

Want to stay up to date with the latest trends in the world of AI, Google algorithms, and modern marketing? Sign up for the Delante newsletter and get a dose of expert knowledge delivered straight to your inbox!

Source: Anthropic News