Anthropic's most powerful models are back online on a limited basis after the company addressed risks that led the government to effectively shut them down, according to a letter from Commerce Secretary Howard Lutnick to Anthropic, seen by Axios.
The abrupt removal of Mythos 5 and Fable 5 created uncertainty within the industry and among allies around the world whose cyber capabilities were stunted by the loss.
Anthropic's engagement with the government has "yielded significant progress," the June 26 letter states.
"In addition, Anthropic has committed to work with the U.S. government on protocols and standards and releases for the Covered Models," the letter states, without elaborating.
The Commerce Department evaluated "diversion risks" presented in the models, and Lutnick determined that appropriate safeguards are in place.
"Accordingly, a license will no longer be required to export, reexport, or in-country transfer (including deemed exports and reexports) the Claude Mythos 5 Model to entities identified in Annex A to this letter and their foreign national employees, or to Anthropic's foreign national employees," the letter states.
The government's June 12 order that prompted Anthropic to take down the models is part of ad hoc regulatory regime.
Export controls remain in place for all organizations not explicitly approved by the administration, and the letter does not change restrictions on Fable 5.
The loosening of restrictions on Anthropic comes after OpenAI's model was subjected to government reviews that the company said were not sustainable.
Meanwhile, in the letter, Lutnick made clear that he can change his mind: "I reserve the right to reevaluate and adjust the scope of license requirements on the Covered Models, should circumstances change," the letter states.
Lutnick also reserved the right to change the list of entities that have access "at any time."
Semafor first reported on the letter.
An Anthropic spokesperson told Axios that the company has received the notice and is now "working to provision the approved set of providers and restore their access to Mythos 5 as quickly as possible."
"We are pleased to see this progress and continue to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again," the spokesperson added.
Lutnick slapped sweeping export controls on Anthropic after Amazon raised jailbreaking concerns, and the company failed to respond seriously, according to the administration.
It's not clear how exactly those concerns were addressed.
There's more regulatory uncertainty ahead, as an August deadline looms for implementation of a cybersecurity executive order. The order calls for federal agencies to create a formal process for assessing AI models' cyber capabilities.
Anthropic has been in discussions with government officials about a formalized policy framework to address national security concerns before models are released, a source familiar with the talks said.
Posted by John3262005
3 Comments
Relevant for the discussion around the US and its relationship regarding AI
Since OpenAI is getting limited release, now Anthropic is getting it too
According to other articles, this is a partially lift of the US ban so that is something
Though in the long run, I don’t know
I wonder if they fixed the problems bought up by Amazon, if they ever did. Or just another way of the Trump administration control on things
The “risks” were always an excuse, what Amazon showed wasn’t a “jailbreak”, companies are just tripping over their shoelaces to suck up to the King.
Supporters of open models models over closed models are being vindicated, and I’m sure the EU is taking note and realizing they can’t depend on American AI companies. Right now the NSA and Friends Of The Trump Admin have full access to Mythos while European governments and companies don’t.
The AI boom is over and it’s the AI companies own fault. They wanted to fear monger about AI capabilities to get the government to implement regulations that would pull the ladder up and insulate them from competition and are now surprised the government is taking them seriously. The end result is in order to do business they’ve given the executive branch the power to destroy their businesses at any time. All a future anti-AI administration (which is likely given how negative public sentiment is on the technology) needs to do is invoke national security and restrict their customers to just the US government; even if they challenge the decision in court their chances of getting an injunction is slim to none and can a company possibly survive if it’s not allowed to sell products while a multi year law suit is in ongoing? Even for the remainder of this administration this sharply limits the market for their most advanced products, and that makes the ROI on developing new models significantly worse. How is any investor supposed to be comfortable investing the amount of capital these companies require going forward when the government can destroy them with the stroke of a pen? How can a company expect a good IPO under these circumstances?