Army’s Ultimatum to Anthropic Over AI Access
Secretary of the Army, Pete Hegseth, has reportedly instructed Dario Amodei, CEO of Anthropic, to provide the military with unrestricted access to the company’s AI models by Friday evening, or face significant consequences. This ultimatum was presented during a tense meeting earlier this week.
During discussions, Hegseth emphasized that the Pentagon would have to either sever its ties with Anthropic, labeling the firm as a risk to the supply chain, or implement the Defense Production Act to compel the company to modify its AI models for military needs without any safeguards.
This standoff marks a notable escalation in the ongoing tensions between the Pentagon and Anthropic regarding AI safety protocols. Defense officials are caught between penalizing the company for its usage constraints and acknowledging their dependence on Anthropic’s advanced AI model, Claude. They conceded that the only reason discussions are ongoing is the military’s immediate need for Anthropic’s technology, which is recognized as cutting-edge.
While Anthropic has suggested that it may alter its usage policies for military applications, it remains hesitant to permit its technology to be used for mass surveillance of American citizens or the development of autonomous weapons that operate without human oversight. Currently, Claude is unique in being approved for the military’s most confidential tasks.
A senior official described the atmosphere of the meeting as tense, though another source portrayed it as friendly, reporting that no voices were raised and that Hegseth actually commended Claude’s performance during his interaction with Amodei.
Hegseth underscored the Pentagon’s position, asserting that private companies cannot dictate operational decisions or object to specific use cases. He referred to the Pentagon’s claims that Anthropic had raised issues with its partner, Palantir, regarding Claude’s implementation in a military operation aimed at Venezuelan leader Nicolás Maduro.
In response, Amodei firmly denied that Anthropic had raised any concerns or discussed such matters with Palantir outside of routine operational communications. He assured that the company’s guidelines have not interfered with Department of Defense operations or affected field personnel.
Following the meeting, Anthropic maintained a diplomatic attitude. A company spokesperson remarked that Amodei expressed gratitude to the ministry and acknowledged Hegseth’s efforts, adding that discussions continue regarding adjustments to usage policies to enable Anthropic’s support for national security missions where its models can be effectively utilized.
The Defense Production Act, which Hegseth hinted at invoking, empowers the president to mandate that private firms prioritize necessary contracts for national defense. This law was previously used to ramp up production of vaccines and ventilators amid the COVID-19 pandemic. However, employing it in such a confrontational context is considered highly unusual. The Pentagon’s goal would be to adapt the model for military purposes without introducing additional safeguards.
The Pentagon is contemplating cutting its contract with Anthropic, which would categorize the firm as a risk to the supply chain. This might compel other contractors within the Pentagon to ensure that Claude is excluded from their operations.
Ending the relationship with Anthropic presents its own set of challenges. The Pentagon would need to find a replacement for Claude quickly, as it is particularly integrated into classified systems. Claude’s military functions have spanned from operations in Venezuela to various administrative tasks throughout the military.
The implications of autonomous weapons and the integration of AI into national security are significant topics that are being explored in a forthcoming book by Winton Hall, a social media director at Breitbart News.
