The Pentagon on Friday chose OpenAI's artificial intelligence models to the detriment of Anthropic, which had refused to open its own without restrictions to the American army in the name of ethics.
Published on: 02/28/2026 - 02:05 Modified on: 02/28/2026 - 10:54
The boss of OpenAI, Sam Altman, announced on Friday February 27 an agreement with the American Department of Defense, which will be able to use its artificial intelligence (AI) models while respecting certain “red lines”.
“Tonight we reached an agreement with the Department of War to deploy our models in their classified network,” Sam Altman wrote on X, using the name given by the Trump administration to his Department of Defense.
Also read: Anthropic against the Trump administration, a decisive standoff over the use of AI by the army
According to Sam Altman, this agreement includes "the prohibition of mass surveillance at the national level and human responsibility in the use of force, including for autonomous weapons systems", i.e. exactly the same red lines as those previously posed by Anthropic so that the Pentagon can use its AI models.
“We will also put in place technical safeguards to ensure that our models behave as they should, which is also what the War Department wanted,” continued the boss of OpenAI, creator of ChatGPT.
This announcement comes a few hours after Donald Trump ordered his administration to “immediately cease all use” of Anthropic’s AI, Claude.
“We don’t need them, we don’t want them, and we won’t work with them anymore,” thundered the American president on his Truth Social platform.
“Their selfishness endangers American lives, our troops and national security,” accused Donald Trump. “The United States will never let a radical, woke left-wing corporation dictate to our great military how to fight and win wars!” he wrote in capital letters.
Defense Minister Pete Hegseth accused Anthropic of "betrayal", castigated the "ideological whims of the tech giants" and banned the company from any direct or indirect collaboration with the American army.
“No amount of intimidation will change our position”
Saying it was “deeply saddened” by this decision, Anthropic said in a statement that “this designation would be both legally unfounded and would set a dangerous precedent for any American company negotiating with the government.” She promised to take legal action.
“No intimidation or sanctions from the War Ministry will change our position,” she further assured.
Anthropic refused Thursday to give in to an ultimatum from Pete Hegseth demanding that it grant the U.S. military unrestricted use of Claude.
“In a limited number of cases, we believe that AI can harm democratic values, rather than defend them,” defended its boss Dario Amodei, who signed a $200 million contract with the American government in June.
“Using these systems for domestic mass surveillance is incompatible with democratic values,” he added.
The 40-year-old leader insists that the most advanced AI systems are not yet so reliable as to give them the power to control deadly weapons – and therefore kill – without human supervision as a last resort.
Fully autonomous weapons “must be deployed with appropriate safeguards, which do not exist today,” said Dario Amodei. “We will not knowingly provide a product that puts U.S. military and civilian personnel at risk.”
Founded in 2021 by OpenAI alumni, Anthropic has always claimed an ethical approach to AI.
At the beginning of 2026, the start-up published a document called "constitution" which details a series of instructions given to Claude to supervise its production. They aim in particular to “prevent inappropriately dangerous actions”.
In announcing OpenAI's agreement with the Pentagon, Sam Altman said he asked the department "to offer these same conditions to all AI companies." “We expressed our strong desire to see things calm down, away from legal and governmental actions, to reach reasonable agreements,” he wrote.
With AFP
Keywords associated with the article
The content you are trying to access does not exist or is no longer available.
