Anthropic wants to grant more organizations access to the Claude Mythos model, but the US government is not on board.
Anthropic and the White House have found a new point of contention. After attempts at cautious rapprochement, both parties now disagree on access to Claude Mythos, the much-discussed security model. Anthropic wants to open its model to more parties, but the White House does not agree.
Ever since its launch a month ago, Anthropic has kept Claude Mythos behind closed doors. This is intended to prevent the model from falling into the wrong hands. For now, 70 approved parties, primarily from the public and technology sectors, have access to the model. Anthropic now wants to admit fifty additional parties.
The White House is reluctant to allow this for reasons beyond security alone. Several US government agencies use the Claude Mythos model. Sources within the government told the Wall Street Journal that Anthropic lacks the computing capacity to handle the increased use of Mythos.
Cautious rapprochement
Anthropic has been at odds with the US government for some time after a dispute over ethical AI use soured its relationship with the Pentagon. The AI company wanted restrictions on the use of its AI models in a military context, such as mass surveillance and the development of autonomous weapons. Defense Secretary Pete Hegseth found those demands so inappropriate that he promptly placed Anthropic on the list of national security risks.
The launch of Claude Mythos brought Anthropic and the White House a bit closer together again. The government was so impressed by the model’s capabilities that it was willing to ignore its own embargo. For now, that government embargo remains in place, despite legal action from Anthropic.
Now the White House wants to interfere with who does and does not have access to Claude Mythos. The government’s desire for control in this matter may not be entirely exaggerated. It appears Anthropic does not have full control over the model’s distribution, after it emerged that ‘unauthorized users’ had found their way to it. These could potentially be hackers from a nation hostile to the United States.