Sam Altman, CEO of OpenAI, has confirmed a strategic agreement with the U.S. Department of Defense (DOD) to deploy its Artificial Intelligence models on the government's classified network. Altman emphasized that both parties share key security principles: the prohibition of national mass surveillance and the requirement for human accountability in the use of force, including autonomous weapon systems. "The Department of Defense reflects these principles in its legislation and policies, and we have integrated them into our agreement," Altman stated on social media, highlighting that AI safety and the equitable distribution of its benefits are fundamental pillars of OpenAI. Altman praised the collaborative attitude of the DOD throughout the negotiations, emphasizing mutual respect and the Pentagon's interest in establishing a partnership that ensures the best outcomes. The agreement includes the implementation of technical safeguards to ensure the proper behavior of the models, a shared priority for both parties. Deputy Secretary of Defense Emil Michael also celebrated the alliance on social media, calling OpenAI "a reliable and stable partner." Michael stressed the importance of having good-faith allies on critical national security issues, especially in the context of AI adoption. Altman announced that FDE will be implemented to support the models and that, for security reasons, they will only be deployed on cloud networks. He also urged the Department of Labor to extend these conditions to all AI companies, arguing that they should be accepted across the sector. He expressed a desire to avoid legal and regulatory disputes, advocating for reasonable agreements. This announcement comes after President Donald Trump's order to ban the use of Anthropic's AI in federal agencies, after the company refused to modify its contracts to allow for mass surveillance or the development of autonomous weapons. Dario Amodei, CEO of Anthropic, reiterated his refusal to accept conditions that he believes could compromise democratic values.