DooDooLamb News
Pentagon strikes classified AI deals with OpenAI, Google, and Nvidia — but not Anthropic
Brief published May 3, 2026 · Original source published May 1, 2026
Original reporting by Emma Roth at theverge.com.
Automated brief. Verify important details at the original source.
Pentagon Builds AI Dream Team, Forgets to Invite the One Company That Actually Cares About Safety
The Department of Defense announced Friday that it has officially welcomed OpenAI, Google, Microsoft, Amazon, Nvidia, Elon Musk's xAI, and startup Reflection into its classified AI club, while conspicuously leaving Anthropic sitting alone at the lunch table like the kid who brings homemade sandwiches to a pizza party. Because nothing says "responsible AI deployment" quite like excluding the company that built its entire brand around not accidentally ending civilization.
The Pentagon's new roster of AI partners reads like a who's who of companies that have never met a user agreement they couldn't make longer or a privacy policy they couldn't make more confusing. OpenAI, fresh off its latest round of "we're definitely still focused on safety" corporate restructuring, gets to bring GPT into America's most sensitive military applications. Google, which once had "Don't Be Evil" as a motto before quietly retiring it like an embarrassing yearbook photo, will presumably help the Defense Department organize its classified files with the same precision it uses to surface your 2003 beach vacation photos at random intervals.
The irony here is thicker than the redacted portions of a CIA report. Anthropic, the company founded specifically because its leadership thought OpenAI was moving too fast and breaking too many things, apparently didn't make the cut for handling the nation's most sensitive military AI needs. This is roughly equivalent to NASA choosing SpaceX, Blue Origin, Virgin Galactic, and a paper airplane startup for a Mars mission while telling the meticulous German rocket engineers they're "probably too careful" for space travel. Anthropic's entire existence is predicated on the radical notion that maybe, just maybe, we should think twice before plugging superintelligent systems into infrastructure that could theoretically turn the planet into a parking lot.
The Pentagon's decision becomes even more fascinating when you consider that Anthropic was previously cleared for classified work, meaning someone in a windowless government building actively decided to revoke their invitation to this particular Silicon Valley prom. Meanwhile, xAI, Elon Musk's latest "I can build it better and faster" vanity project, somehow made the list despite being roughly as old as a TikTok trend and about as battle-tested as a paper submarine. The same Elon Musk who has spent the better part of two years turning Twitter into a case study in how to alienate users, advertisers, and basic functionality is now trusted with classified AI applications. The man who thought "X" was a better name than "Twitter" will help determine how artificial intelligence integrates with national defense strategy.
Google's inclusion is particularly rich given the company's track record with military contracts. This is the same Google that faced massive internal protests over Project Maven, its AI-powered drone targeting system, ultimately backing out amid employee revolts and ethical concerns. But apparently those concerns evaporated faster than a Stadia console's relevance once the Pentagon started writing checks with more zeros. Google's motto may no longer be "Don't Be Evil," but "Don't Pass Up Government Contracts" seems to be working just fine as a replacement.
The presence of Microsoft and Amazon surprises exactly no one who has watched the cloud wars unfold over the past decade. Both companies have been circling government contracts like sharks around a wounded seal, with Amazon's AWS and Microsoft's Azure already powering vast swaths of federal infrastructure. Adding AI capabilities to their existing government relationships is less a strategic pivot and more a natural evolution, like adding a spoiler to a race car that was already going 200 miles per hour. These companies have spent years convincing government agencies that the cloud is safe, secure, and definitely won't be used to train AI models on sensitive data (wink, nudge, trust us).
Nvidia's inclusion makes perfect sense from a hardware perspective, considering they've effectively cornered the market on AI training chips the way De Beers once controlled diamonds. Every AI model worth its computational salt runs on Nvidia GPUs, making their partnership with the Pentagon less about AI strategy and more about acknowledging basic supply chain reality. Excluding Nvidia from AI discussions would be like trying to build a car without mentioning steel or attempting social media without manufactured outrage.
But the real star of this government roster has to be Reflection, a startup so new and unproven that including them feels like the Pentagon decided to hedge their bets by throwing a dart at a Silicon Valley phone book. Reflection's claim to fame appears to be existing at exactly the right moment when the Defense Department needed to pad their AI partner list, which is roughly equivalent to being selected for the Olympic swimming team because you happened to be standing near the pool when they needed an eighth swimmer.
The practical implications of these partnerships remain as murky as a Pentagon budget briefing. These companies will presumably help the military process intelligence, coordinate logistics, and make strategic decisions using AI systems trained on datasets that may or may not include the entirety of human knowledge scraped from the internet without permission. The same technology that occasionally insists that glue is an acceptable pizza topping will now help analyze threats to national security. What could possibly go wrong with letting ChatGPT's cousins loose on classified intelligence reports?
For developers and tech workers watching this unfold, the message is clear: safety-first approaches and ethical considerations are nice to have until someone with a bigger budget and more urgent timeline comes knocking. The Pentagon's AI partner selection reads like a case study in choosing the companies most likely to deliver fast results over those most likely to deliver safe ones.
The exclusion of Anthropic sends a particularly chilling message about where the industry's priorities actually lie when the rubber meets the classified road.