Washington D.C. – OpenAI is in discussions with the U.S. Department of War regarding potential agreements for the use of its artificial intelligence models and tools, according to a source with knowledge of the talks. The move comes amid a public dispute between the Pentagon and Anthropic, another leading AI developer, and a recent directive from President Trump to cease federal contracts with Anthropic. The emerging negotiations highlight the complex relationship between the government and the rapidly evolving AI industry, particularly concerning national security applications and ethical considerations surrounding artificial intelligence in autonomous weapons.
The potential deal with OpenAI was revealed during an all-hands meeting with employees on Friday, where CEO Sam Altman indicated the government is willing to allow OpenAI to establish its own “safety stack”—a layered system of technical, policy, and human controls designed to govern the use of its AI models. Crucially, Altman stated the government has signaled it would not compel OpenAI to deploy AI in ways that violate the company’s pre-defined “red lines,” which include prohibiting the use of AI for autonomous weapons systems, domestic mass surveillance, and critical decision-making processes. This concession represents a significant shift in the government’s approach, as it previously demanded Anthropic allow for “all lawful purposes” when deploying its AI technology.
The Anthropic Fallout and Shifting Dynamics
The situation with OpenAI unfolds against the backdrop of a dramatic breakdown in relations between the Pentagon and Anthropic. Secretary of War Pete Hegseth and Anthropic CEO Dario Amodei engaged in a public disagreement over limitations Anthropic sought to impose on the use of its Claude AI model. Anthropic refused to remove safeguards preventing its technology from being used for domestic mass surveillance or in fully autonomous weapons, even facing the threat of losing a potential $200 million contract. According to a source at OpenAI, the relationship between Anthropic and the government deteriorated further due to perceived offenses from Amodei, including the publication of blog posts that reportedly upset Department of War leadership.
Anthropic had previously been the sole large commercial AI provider with models approved for use within the Pentagon’s classified systems, operating through a partnership with Palantir. The dispute underscores the fundamental tension between the desire for advanced AI capabilities within the military and the ethical concerns surrounding the technology’s potential misuse. OpenAI, which shares Anthropic’s “red lines” regarding military applications of AI, now finds itself navigating the same complex landscape.
Trump Administration Intervenes
Adding another layer of complexity, President Trump announced on Friday that the federal government would immediately cease all work with Anthropic. “I am directing every federal agency in the United States government to immediately cease all use of Anthropic’s technology. We don’t need it, we don’t seek it, and will not do business with them again!” Trump stated in a post on Truth Social. Agencies using Anthropic’s Claude models have been given a six-month phase-out period. This decisive action signals a strong stance against AI developers perceived as resistant to government demands.
Concerns Over Surveillance and National Security
During the OpenAI all-hands meeting, company leaders acknowledged the challenges surrounding potential government surveillance applications of AI. Even as expressing concerns about the potential for AI-driven surveillance to threaten democratic principles, they similarly recognized the reality that governments require intelligence gathering capabilities, particularly in the context of international threats. References were made to threat intelligence reports indicating that China is already utilizing AI models to target dissidents abroad, highlighting the competitive pressures driving the demand for advanced AI technologies within the national security apparatus.
OpenAI’s proposed “safety stack” would limit deployment to cloud environments, avoiding “edge systems” like aircraft and drones, and retain company control over the implementation of safeguards and model deployment. This approach aims to balance the government’s need for advanced AI capabilities with OpenAI’s commitment to responsible AI development. Sasha Baker, head of national security policy at OpenAI, and Katrina Mulligan, who leads national security for OpenAI for Government, were also present at the all-hands meeting, according to the source.
The negotiations between OpenAI and the Department of War represent a pivotal moment in the evolving relationship between the government and the AI industry. The outcome of these discussions will likely set a precedent for future collaborations and shape the ethical boundaries of AI deployment within the military. The situation also underscores the growing importance of establishing clear “red lines” and robust safety mechanisms to mitigate the risks associated with increasingly powerful AI technologies. Further updates on the negotiations are expected in the coming weeks, as both sides work to finalize the terms of a potential agreement.
This is a developing story. Readers seeking support related to concerns about surveillance or privacy can find resources at the Electronic Frontier Foundation (https://www.eff.org/).
What do you think about the government’s approach to AI regulation? Share your thoughts in the comments below.
