Sam Altman was at the Vanity Fair Oscar party. It was supposed to be champagne and celebrity sightings. Instead, he got cornered about weapons contracts.
The OpenAI CEO found himself confronted by independent journalist Sam Sacks at the exclusive post-ceremony gathering in early March, pressed on his company’s expanding relationship with the U.S. Department of Defense and intelligence agencies. The exchange, captured on video and reported by Futurism, showed Altman visibly uncomfortable as Sacks asked pointed questions about OpenAI’s military partnerships — partnerships that would have been unthinkable just two years ago, when the company still maintained a blanket prohibition on military and warfare applications of its technology.
Altman’s response was telling. Rather than defending the contracts directly, he offered a vague assurance that OpenAI’s work with the Pentagon involved “cybersecurity” and wasn’t being used to “hurt people.” He then attempted to redirect, suggesting Sacks talk to OpenAI’s policy team. It was the kind of deflection that plays well in boardrooms but lands awkwardly when you’re in a tuxedo holding a cocktail glass.
The confrontation matters because it crystallizes a tension that has been building inside OpenAI and across the broader artificial intelligence industry for more than a year. The company that began as a nonprofit research lab dedicated to ensuring AI benefits all of humanity has been on an aggressive march toward commercialization, and its pivot toward defense and intelligence work represents perhaps the starkest departure from its founding ethos.
OpenAI quietly removed its prohibition on military use cases from its usage policies in January 2024. The change didn’t come with a press conference or a blog post explaining the philosophical reasoning. It was simply edited out, noticed by sharp-eyed observers and flagged in subsequent reporting. At the time, OpenAI spokesperson Niko Felix told media outlets that the company had updated its policies to be “clearer” and “more readable,” and that it still prohibited the use of its technology to develop weapons or cause harm. But the removal of explicit military-use language opened a door that OpenAI has since walked through with increasing confidence.
And walk through it they have. OpenAI struck a deal with the Pentagon for cybersecurity applications. It partnered with Anduril Industries, the defense technology company founded by Palmer Luckey, to work on counter-drone systems. It has engaged with U.S. intelligence agencies. Each new contract has been framed as defensive in nature — protecting networks, countering threats, supporting national security in ways that don’t involve lethal applications. The framing is careful. Deliberate.
But critics aren’t buying it.
The concern, articulated by Sacks during the Oscar party encounter and echoed by numerous AI ethicists and former OpenAI employees, is that the line between defensive and offensive military AI is inherently blurry. Cybersecurity tools can be used for surveillance. Counter-drone systems involve targeting decisions. And once a company’s technology is integrated into military command structures, the company has limited control over how it’s ultimately deployed.
This isn’t hypothetical hand-wringing. The history of dual-use technology is littered with examples of tools built for one purpose being repurposed for another. GPS was a military technology before it helped you find the nearest coffee shop. The internet itself was a Defense Department project. The difference with AI is the speed at which capabilities can be redirected and the opacity of the systems once deployed.
OpenAI’s transformation from cautious nonprofit to aggressive commercial entity has been one of the most dramatic corporate metamorphoses in recent tech history. The company’s ongoing restructuring — converting from its unusual capped-profit structure to a full for-profit corporation — has drawn scrutiny from state attorneys general, former board members, and co-founder Elon Musk, who has filed lawsuits attempting to block the transition. The defense contracts add another layer to this story, suggesting that OpenAI’s commercial ambitions extend well beyond consumer chatbots and enterprise software.
The financial incentives are enormous. The U.S. defense budget for fiscal year 2025 exceeds $880 billion, and the Pentagon has been aggressively courting Silicon Valley AI companies. The Department of Defense’s Chief Digital and Artificial Intelligence Office has been working to accelerate AI adoption across military branches. For a company reportedly seeking valuations north of $300 billion, defense revenue represents a massive and reliable revenue stream — one backed by the full faith and credit of the United States government.
OpenAI isn’t alone in this pivot. Google, which famously faced an employee revolt over its Project Maven drone imagery work in 2018, has since quietly rebuilt its defense business. Microsoft, OpenAI’s largest investor and closest partner, has deep longstanding Pentagon relationships through its Azure Government cloud services and a $10 billion JEDI contract (later restructured). Amazon Web Services runs classified cloud infrastructure for the intelligence community. Palantir and Anduril have built entire businesses around defense AI.
So the industry trend is clear. What makes OpenAI’s case distinctive is the velocity of the reversal and the gap between its public rhetoric and its commercial trajectory.
Altman has spent years cultivating an image as a thoughtful steward of potentially dangerous technology. He testified before Congress about the need for AI regulation. He’s spoken eloquently about existential risk. He’s positioned OpenAI as the responsible actor in a field full of reckless competitors. The defense contracts complicate that narrative considerably — not because defense work is inherently wrong, but because Altman has been reluctant to engage with the tension honestly and publicly.
The Oscar party confrontation illustrated this reluctance perfectly. When pressed, Altman didn’t make the case that national security applications of AI are necessary and appropriate. He didn’t argue that American AI companies have a patriotic obligation to support the defense establishment. He didn’t even make the pragmatic case that if OpenAI doesn’t do this work, Chinese competitors will. He just said it was about cybersecurity and suggested the journalist talk to someone else.
That’s not leadership. That’s avoidance.
Former OpenAI employees have been among the most vocal critics of the company’s direction. Several who left in 2024 signed open letters warning about the erosion of safety culture at the company. Others have spoken to reporters on background about what they describe as a systematic deprioritization of safety research in favor of product development and commercial partnerships. The departure of key safety researchers — including co-founder Ilya Sutskever and safety team lead Jan Leike — lent credibility to these concerns.
The defense partnerships raise a specific version of this broader safety question: Who provides oversight when AI systems are deployed in military contexts? OpenAI’s terms of service and acceptable use policies are one thing when applied to a college student using ChatGPT to write an essay. They’re something else entirely when applied to systems integrated into Pentagon operations. The company has said it maintains the right to audit and restrict use of its technology, but the practical reality of telling the Department of Defense “no” once contracts are signed and systems are embedded is far more complicated than any policy document suggests.
There’s also the question of what OpenAI’s nonprofit board — which still technically oversees the organization during its transition period — thinks about all of this. The board was reconstituted after the chaotic November 2023 episode in which directors briefly fired Altman before reversing course under pressure from employees and Microsoft. The current board includes figures like Bret Taylor, Larry Summers, and retired Admiral Michael Mullen. Mullen’s presence is itself a signal: the former Chairman of the Joint Chiefs of Staff presumably wasn’t added to the board because of his expertise in natural language processing.
The broader geopolitical context can’t be ignored either. The AI arms race between the United States and China has become a central preoccupation of Washington policymakers. Export controls on advanced chips, restrictions on Chinese AI companies, and bipartisan support for American AI dominance have created an environment where defense partnerships aren’t just commercially attractive — they’re politically advantageous. An AI company that works with the Pentagon is an AI company that has friends on Capitol Hill. And OpenAI needs friends on Capitol Hill as it faces regulatory scrutiny over its corporate restructuring, its data practices, and its market dominance.
This political calculus helps explain why Altman has been reluctant to engage with critics on the defense question. Defending the contracts too publicly risks alienating the progressive tech workers and academic researchers who form part of OpenAI’s talent pipeline. But distancing himself from the contracts risks alienating the policymakers and defense officials whose goodwill he needs. So he splits the difference: the contracts exist, but he’d rather not talk about them at parties.
The problem with this approach is that it satisfies nobody. AI safety advocates see the defense work as confirmation of their worst fears about OpenAI’s trajectory. Defense hawks see Altman’s discomfort as evidence that Silicon Valley can’t be trusted with serious national security work. And the general public — to the extent it’s paying attention — sees a CEO who talks about safety while signing military contracts, which looks a lot like hypocrisy regardless of the underlying merits.
What would honest engagement look like? It would start with Altman making a clear, public case for why OpenAI’s defense work is consistent with its mission. Not “talk to our policy team.” Not vague gestures toward cybersecurity. A real argument, with specifics about what the technology is being used for, what safeguards are in place, and what red lines the company won’t cross. It would include transparency about the decision-making process that led to removing the military-use prohibition. It would involve acknowledging the legitimate concerns of critics rather than dismissing them.
None of this has happened. And the longer it doesn’t happen, the more the gap between OpenAI’s stated values and its actual behavior becomes the story.
The AI industry is entering a period where the decisions made by a handful of companies will shape how this technology intersects with military power, surveillance, and state authority for decades. These decisions deserve public debate, not cocktail-party deflections. Sam Altman built OpenAI on the promise that artificial intelligence would be developed responsibly and for the benefit of humanity. Whether Pentagon contracts are consistent with that promise is a legitimate question. It deserves a real answer.
He owes us more than “talk to our policy team.”
WebProNews is an iEntry Publication