Altman Admits “Bad Optics” but Says Pentagon Deal Prevents Wider Industry Crackdown
OpenAI CEO Sam Altman commented extensively on Saturday on his company’s controversial deal with the U.S. Department of Defense. In a question-and-answer session on X.com, he acknowledged that the agreement appeared rushed, but defended it as a necessary step to de-escalate a conflict between the government and the AI industry.
Background: Dispute between Pentagon and Anthropic
The deal came about after negotiations between the Department of Defense (formerly the War Department) and competitor Anthropic failed. The Pentagon subsequently threatened to classify Anthropic as a “risk to national security in the supply chain.” Shortly thereafter, OpenAI reached an agreement with the agency on collaboration in classified areas.
Altman emphasized that OpenAI had long planned only non-classified work with the Department of Defense. Only this week had the situation escalated, prompting swift action.
What the Deal Covers and What Remains Prohibited
According to OpenAI, the company has defined three central red lines that must not be crossed:
- No use of the technology for mass domestic surveillance
- No use for controlling autonomous weapons systems
- No use for automated high-risk decisions (such as social credit systems)
To enforce these boundaries, OpenAI employs a multi-layered approach. AI models will be provided exclusively via the cloud, not on edge devices, which is intended to prevent integration into weapons systems. OpenAI retains full control over its security systems and will involve its own engineers and security researchers in work with the Pentagon.
“We will shut down the technology in the very unlikely event that the rules are violated. But we believe that the U.S. government is an institution that does its best to comply with laws and policies,” Altman explained.
Defense Against Criticism
When asked why OpenAI concluded the deal so quickly, Altman responded: “It was definitely rushed, and the optics don’t look good. We really wanted to de-escalate.” He hopes the agreement will lead to a relaxation of tensions between the Department of Defense and the industry. If so, OpenAI will “look like geniuses”; if not, they will continue to be characterized as rushed and reckless.
Altman also expressed understanding for the Pentagon’s position. On one hand, the AI industry tells the government that the technology is crucial for geopolitical conflicts and that China is ahead. On the other hand, the industry refuses to cooperate and views the agency as “somehow evil.” In that situation, he himself would not respond well, Altman said.
Differences from Anthropic’s Position
OpenAI emphasizes that its own contract contains more safeguards than previous agreements, including Anthropic’s original agreement. While Anthropic focused on specific contractual prohibitions, OpenAI relies on technical security measures and the direct involvement of experts.
Altman also stressed that OpenAI secured the same conditions being offered to all AI companies during negotiations. He hopes that Anthropic and other labs will also accept these conditions.
According to OpenAI, the deal is worth only a few million dollars and is “completely insignificant” compared to the company’s over $20 billion in revenue. The company is doing this not for the revenue, but because it is the right thing for the country, a company spokesperson said.
Criticism of Anthropic’s Classification
Despite its own deal, Altman sharply criticized the Pentagon’s threat against Anthropic. Enforcing the classification as a security risk would be “very bad for our industry and our country and obviously for their company.” He hopes the Department of Defense will reverse this decision, even if OpenAI faces criticism for it.
“I don’t think unelected leaders of private companies should have as much power as our democratically elected government. But I think we need to help them,” Altman said.
