The Trump administration and Anthropic have reached a deal to restore access to Fable, the company’s latest general-access AI model.
According to the Wall Street Journal, the agreement ends a two-and-a-half-week shutdown that left the AI industry twitching and showed how far the White House now wants its fingers in model releases.
Under the deal, Anthropic will fix the workarounds Amazon researchers used to dodge Fable’s safeguards, which it was pretty much going to do anyway.
US commerce secretary Howard Lutnick said on X that Fable, a public version of Anthropic’s more powerful Mythos model, could carry out cyberattacks without proper guardrails.
“Over the past two weeks, we have worked closely with Anthropic to analyse and approve Fable 5 to ensure alignment across the US government and strengthen America’s leadership in AI,” Lutnick said.
Anthropic said it is working with Amazon, Microsoft and Google on a consensus framework for judging jailbreaks and deciding how AI developers should respond.
The outfit admitted it was “probably impossible” to make any model jailbreak-proof, which is not the sort of thing that calms nervous security types.
Anthropic chief compute officer Tom Brown led recent talks with the Commerce Department and other agencies to convince them Fable was safe enough for public use.
After the agreement, Anthropic said it would begin restoring access on Wednesday, ending an unprecedented shutdown for a leading US model maker.
“We’re grateful to our users for their patience, and to everyone who worked with us on redeploying,” the company said.
Anthropic said it had implemented a new safeguard after working closely with the government on the behaviour flagged by Amazon researchers.
According to the company, Amazon’s guardrail-dodging trick now fails about 99 per cent of the time.
For the remaining cases, Anthropic said the outputs were not considered risky because they contained public information or would not help a wrong ’un. The company had previously said risky prompts sent to Fable would be redirected to a less capable model.
Administration officials said the aim was to fix the vulnerability and create a process to stop similar messes from happening again.
All this may have given Anthropic’s rivals a handy breather. OpenAI, maker of ChatGPT, has products with similar capabilities that are being released on a limited basis at the government’s instruction. Google and Elon Musk’s AI company are competing in the same space.
