Anthropic and US government in talks again about cooperation

The US government is talking again with the AI manufacturer Anthropic, which it had sanctioned. It was recently classified by the Pentagon as a “supply chain risk” because it did not want to support the use of its technology for autonomous US weapons systems and mass surveillance of US citizens.

The portal Axios reports, citing several sources from the White House, about the meeting between Anthropic CEO Dario Amodei and US Treasury Secretary Scott Bessent and Chief of Staff Susie Wiles. The background to the meeting was reportedly the provision of Anthropic's new AI Claude Mythos Preview to selected institutions and authorities. Anthropic considers a general release of Claude Mythos Preview to be too dangerous. The capabilities of this new model are currently causing massive cybersecurity concerns among companies and governments worldwide; the BSI also expects major “upheavals” in IT security.

Pressure on the US government apparently too high

The US government apparently felt compelled by developments to resume talks with Anthropic – despite the lawsuit that Anthropic filed against the Pentagon's actions. Both sides wanted to disregard this dispute in the current discussion; it was not a topic. Instead, the focus was on future issues, particularly how other government agencies should handle Claude Mythos Preview.

Wiles reportedly emphasized in the meeting that a constructive relationship with Anthropic is important despite ongoing legal proceedings. According to Axios, the discussion concerned the security of the source code, the company's decision-making processes, and how to handle the release of new models.

Interest in cooperation

Several agencies, including the Treasury Department, reportedly showed interest in cooperation. Currently, Anthropic's classification as a security risk would prohibit this, but since Anthropic has sued against it, these are initially suspended. On the one hand, the US government – like many others – fears that advanced AI tools like Mythos could be misused by cybercriminals to attack critical systems. On the other hand, companies and authorities could use Mythos to strengthen their cybersecurity before criminals gain access.

Scott Bessent reportedly wanted to ensure with his participation in the meeting that “all parties are pulling together” and emphasized in the conversation that the government plays an important role despite the private sector.

The US government subsequently described the meeting as “productive and constructive.” The current talks follow several other discussions by the US government with tech companies, banks, and other companies regarding Anthropic's new AI Claude Mythos Preview.

(nen)