US Orders Anthropic to Block Foreign Access to Fable 5

The US government ordered Anthropic PBC to disable access to its Fable 5 AI model for all foreign nationals, people familiar with the matter said. The order came after officials discovered researchers could jailbreak – bypass guardrails on – the model Anthropic had released just days earlier.

Anthropic personnel are now in talks with Treasury Secretary Scott Bessent and other officials about the specific security concerns, the people said. Bessent has warned Wall Street about frontier models such as Anthropic's Mythos, a platform designed to find software flaws faster than humans.

The move is the most aggressive US incursion into an AI company's operations to date. It arrives as Anthropic, valued at more than $900 billion, and rivals like OpenAI, Google and Meta Platforms prepare for public listings.

Which AI developers feel the heat

The order threatens to set a precedent for how Washington controls advanced AI. A June executive order from President Trump said the US would not create a licensing regime for AI models. This action does exactly that, though the administration has not called it one.

Google and Meta each field frontier models. If Washington can force Anthropic to block foreign access, it can do the same to their models. The risk extends beyond Fable 5. The government's fear is that adversaries might steal model weights – the core file that defines an AI's capabilities. Senator Jim Banks, a Republican from Indiana, recently asked officials to protect those weights from China or other rivals.

“US frontier models are increasingly treated as strategic assets, with access tightly controlled and shaped by national security considerations,” said Gary Tan, a portfolio manager at Allspring Global Investments. That “dynamic is likely to persist as China continues to lag the US in compute.”

The Stakes for Future Model Releases

Anthropic warned that a heavy-handed response could “halt all new model deployments for all frontier model providers.” The company has cautioned about Mythos's cyber capabilities, and White House officials had earlier created a pathway for US agencies to use it to patch their own systems.

The debate unfolds against a race with China. Alibaba Group and DeepSeek are narrowing the gap with US rivals in performance and efficiency. Stefanie Kam, an assistant professor at Nanyang Technological University in Singapore, said Washington is “racing to use export controls to rein in frontier AI risks.” If the directive is narrowly targeted, US firms can adapt. “If it’s sweeping, it risks pushing innovation offshore while China advances,” she said.

Politics aside, Silicon Valley will resist. From Meta to OpenAI, the largest developers are betting their futures on this technology. Yet testing every possible capability of a large model is impossible, leaving vague threats and constant scrutiny.

“There may be important capability thresholds – such as AI systems that can automate large amounts of AI research and development in short periods of time – that pose new kinds of risks,” Banks wrote in a letter to Bessent, National Cyber Director Sean Cairncross and Defense Secretary Pete Hegseth. “Industry experts have likewise raised serious concerns about the difficulty of keeping such systems under human control.”

The order marks a reversal of decades of US policy that spread cutting-edge technology abroad. Now, like curbs on chip exports to China, the US is keeping advanced AI at home.