Anthropic heads to DC to try to unban newest model

Worse than flying near a crying baby is sitting on a red eye surrounded by a group of nerds discussing the existential risk of AI. Some air passengers might’ve suffered that predicament as Anthropic dispatched senior staff to DC this week to meet with government officials. The company is hoping to convince the administration to reverse a ban issued Friday on foreigners using its powerful Fable 5 model, which prompted Anthropic to cut off all access to the model.

Anthropic debuted Fable 5 last week, claiming it was a safer version of its unreleased Mythos model—which it says has potent cybersecurity capabilities that can be weaponized by hackers. But it took Fable down on Friday after the Trump administration restricted its use to Americans only, citing national security and findings that its safeguards could be bypassed by bad actors.

Now, Anthropic is negotiating with administration officials to bring Fable back online. “Both parties are working quickly to get this resolved,” an Anthropic spokesperson told the WSJ yesterday. “This is part of our ongoing commitment to working alongside the administration toward our shared goal of protecting US critical infrastructure and the US lead in cyber defense.”

Fabled dispute

The Trump administration’s decision came after Amazon (an Anthropic investor) reportedly told officials that it had found ways to jailbreak Fable to make it a potential asset for hackers:

• Officials told Politico they restricted Fable after Anthropic failed to address safety concerns in a timely manner.
• An Anthropic insider countered that it was given a 90-minute deadline to voluntarily take the model offline “with no details on the actual threat,” per Politico. The company claims that the jailbreaking threat applies only in narrow cases and that the risk is about the same as similarly powerful tools from rival OpenAI and some Chinese models.

Meanwhile, a group of cybersecurity pros signed an open letter to the government yesterday supporting Anthropic, claiming that it’s important for security experts to have access to models like Fable 5 to help them fortify systems against attacks.

It’s not Anthropic’s only DC business…the company is also currently suing the Defense Department seeking a reversal of its designation as a supply chain risk.

The issue is bigger than Anthropic: Critics said that the government’s decision to ban Fable didn’t follow a transparent procedure and could have a chilling effect on other AI labs. However, some commentators said it was natural that the government was regulating the model after Anthropic spent weeks touting how dangerously powerful its AI can be.—SK