Pentagon Tests Google and OpenAI AI Models After Anthropic Fallout

The US Department of Defense is testing new artificial intelligence models from Google and OpenAI after ending a major deal with Anthropic. The Pentagon designated Anthropic as a "supply chain risk" weeks ago, following disagreements over the use of its Claude AI models. The Pentagon signed a new agreement with OpenAI just hours after the fallout with Anthropic, according to Bloomberg.

Key Highlights

• Pentagon ended deal with Anthropic over AI use restrictions.
• US military now tests AI models from Google and OpenAI.
• 25 power users evaluate alternatives across five military commands.
• Anthropic challenges supply chain risk label in court.
• Officials seek multiple AI providers to avoid overreliance.

Pentagon Shifts AI Partnerships

In March, three days after Defense Secretary Pete Hegseth labeled Anthropic a supply chain risk, the Pentagon began evaluating alternative AI models. The department is working with 25 "power users" to test these models, seeking replacements for Anthropic's Claude. The new models are being tested on a digital platform separate from the Maven Smart System, which previously relied on Claude for classified operations.

The Pentagon had used Anthropic's Claude in the Maven Smart System, which supports classified missions, including operations against Iran. Reports have linked the use of AI to a US strike on an Iranian school in February that resulted in 165 deaths. The fallout with Anthropic followed the company's refusal to meet two Pentagon conditions: allowing use for all lawful purposes and permitting mass domestic surveillance in the US. Anthropic insisted that its AI could not be used for mass domestic surveillance or certain other applications, leading to the breakdown of the agreement.

New AI Models Under Evaluation

After Anthropic withdrew, the Pentagon gave itself six months to phase out use of the company's products. OpenAI CEO Sam Altman later described the new deal with the Pentagon as "rushed." The Pentagon is now testing AI models from companies such as Google and OpenAI across five military theatre commands worldwide. The evaluation involves "power users" who are assessing the capabilities of these alternative models.

Officials have reported less resistance from users than expected, even though some preferred Anthropic's tools. The Pentagon aims to ensure operators have access to multiple AI models, preventing any single company from gaining leverage over military operations. Defense Undersecretary Emil Michael told Bloomberg that rival developers are expected to release models matching Anthropic's capabilities every month or two.

Ongoing Legal and Security Concerns

Anthropic has challenged its supply chain risk designation in court, arguing that the label is typically reserved for foreign adversaries such as Huawei. Despite some signs of de-escalation between the White House and Anthropic, Pentagon officials have not indicated any plans to restore the partnership. Security concerns remain, especially as new AI models are described as potential threats to existing cyber defenses.

The Pentagon's shift highlights the growing importance of AI in military operations and the challenges of balancing security, ethical use, and supplier relationships. The department continues to evaluate AI options to maintain operational flexibility and reduce reliance on any single provider.