US Govt Suspends Access to Anthropic’s AI Models amid Security Concerns

TL/DR –

AI lab Anthropic recently suspended its latest models, Fable 5 and Mythos 5, in response to a US government directive limiting use to US nationals. The company has had ongoing conflicts with the Trump administration over AI regulation and export policy, and declined requests to let the Pentagon use its models for surveillance and weapons systems. These models, which utilise machine learning methods and large language models, are complex and their behaviours are not fully understood, making regulation difficult.

An AI Lab’s Recent Shutdown of Models Under US Directive

On June 12, AI lab Anthropic discontinued access to its advanced Claude models, Fable 5 and Mythos 5, due to a directive by the US government.

Identified as frontier models, Mythos 5 was initially withheld from release due to its hacking proficiency while Fable 5, equipped with cybersecurity safeguards, was launched and promptly suspended last week.

Anthropic’s Conflict With the Trump Administration

The Trump administration and Anthropic have locked horns since early 2025 over issues like AI regulation, semiconductor export policy, and the refusal of Anthropic to allow Pentagon use of their models for surveillance and automation.

Jailbreak Fears

Anthropic suspects the US government’s shutdown directive was due to a potential jailbreak, or a circumvention of Fable 5’s safeguards meant to prevent misuse. It’s believed the safeguards could be bypassed to extract data for cyberattacks.

Anthropic also addressed the existence of an online community, referred to as the Undersphere, aiming to bypass AI safeguard systems.

Opaque AI Models: A Regulatory Challenge

Due to their opacity, AI models are difficult to regulate. Governments lack the necessary data and resources to independently evaluate these models. The Trump administration recently issued an executive order requesting developers to submit their models for review before release.

AI Governance Framework Required

Considering AI’s unpredictability and power, a global, participatory governance framework capable of predicting and addressing system failures is necessary. Unfortunately, current US administration has not demonstrated capacity for such a system.

Written by Francesco Bailo, a senior lecturer at the University of Sydney, this article is republished from The Conversation under a Creative Commons license. Read the original article here.

—
Read More US Political News