The Trump administration has developed stringent guidelines for artificial intelligence (AI) contracts in the civilian sector, demanding that companies allow for "any lawful" use of their AI models. This move comes amid a conflict between the Pentagon and Anthropic, as reported by the Financial Times on Friday.
On Thursday, the Pentagon classified Anthropic as a "supply-chain risk," prohibiting government contractors from utilizing the AI firm's technology in projects for the US military. This decision follows a prolonged disagreement over the company's insistence on implementing safeguards that the Defense Department deemed excessive.
A preliminary version of the guidelines, as reviewed by the FT, stipulates that AI companies seeking government contracts must provide the US with an irrevocable license to utilize their systems for all legal purposes.
According to the Financial Times, the General Services Administration (GSA) is set to enforce these regulations for civilian contracts as part of a broader initiative to enhance the procurement of AI services. These measures are reportedly aligned with those under consideration by the Pentagon for military contracts.
Josh Gruenbaum, commissioner of the Federal Acquisition Service, a GSA subsidiary facilitating software procurement for the federal government, expressed via email to Reuters the necessity of severing ties with Anthropic. He mentioned that terminating Anthropic's OneGov deal was in compliance with the President's directive, effectively ending their availability to the Executive, Legislative, and Judicial branches through GSA's pre-negotiated contracts.
Requests for comment from Reuters to the White House went unanswered at the time of reporting.
The GSA draft also mandates that contractors refrain from intentionally embedding partisan or ideological biases into the data outputs of AI systems. Furthermore, companies are required to disclose whether their models have been altered or configured to comply with any non-US federal government or commercial regulatory frameworks.
