Anthropic suspends new AI tools over US government security concerns
Anthropic says it has disabled its new Fable 5 and Mythos 5 AI tools after US authorities raised security concerns.
Intelligence analysis by GPT-5.4 Mini
Anthropic pulled back its newest AI models days after release, saying US officials ordered a suspension affecting foreign nationals and forcing the company to disable the tools for all customers. The company says the concern appears tied to a possible jailbreak method, not a named vulnerability.
Anthropic made a very strong robot brain, then had to turn it off after the government worried it might be used the wrong way. It is like a new car that can go very fast, but officials worry the brakes are not ready yet.
Analysis
What happened
Anthropic said it has suspended access to its newly released AI models, Claude Fable 5 and Mythos 5, after US authorities raised security concerns soon after the public launch. The company said it was ordered to stop foreign nationals from using Fable 5, and that the practical result was to disable both models for all customers.
What the company says
Anthropic said US national security officials had not pointed to a specific flaw. Instead, the company believes the government learned of a method that could bypass the model's protections, often described as jailbreaking. In Anthropic's account, a demonstration of that method led to the discovery of a small number of already known minor weaknesses. The company argued those issues were relatively simple and said other public models could also uncover them without needing the bypass.
Wider context
The article says the launch had already worried finance, technology, and government figures because the model was tested privately before release and Anthropic had previously described it as too powerful for public use. The company says it provided early access to a small number of organisations to examine whether the system could be misused for hacking or cyber exploitation.
Why the debate is growing
Gina Neff of Queen Mary University London told the BBC that limiting access could also limit safe testing and collaboration with governments. She said the field is entering uncharted territory. The BBC also reports that the UK government's AI Security Institute found the model could exploit defences and systems 73% of the time in its tests.
The story comes amid a separate legal fight between Anthropic and the Trump administration over a Pentagon designation that called the company a supply chain risk and over an order affecting government use of its tools.
Key points
- Anthropic says it suspended Claude Fable 5 and Mythos 5 after US authorities raised security concerns.
- The company says it was ordered to stop foreign nationals from using Fable 5, which forced a wider shutdown for all customers.
- Anthropic says officials did not identify a specific flaw, but may have learned of a jailbreak method.
- The article says the UK AI Security Institute found the model could exploit defenses and systems 73% of the time in tests.
- The company is also involved in a separate lawsuit with the Trump administration over government use of its AI tools.
If the pause leads to better testing, Anthropic could strengthen the models before wider use resumes. The extra scrutiny could also help governments and companies build safer ways to use advanced AI.
If the security concerns remain unresolved, the models could stay restricted or face tighter rules. The dispute could also make it harder for researchers and governments to test powerful AI systems together.
