The Pentagon on Friday signed agreements with seven main synthetic intelligence corporations to run their expertise on categorised navy networks. The Warfare Division named the businesses as SpaceX, OpenAI, Google, Nvidia, Reflection, Microsoft, and Amazon Internet Providers. “Collectively, the Warfare Division and these strategic companions share the conviction that American management in AI is indispensable to nationwide safety,” the Pentagon mentioned in its official assertion Friday.The offers authorize the businesses to deploy AI on Impression Stage 6 and IL7 networks—the federal government’s highest-tier categorised environments, used for all the pieces from secret intelligence evaluation to delicate nationwide safety knowledge. The Pentagon mentioned the purpose is to make the US navy an “AI-first preventing power.”One identify lacking from the checklist: Anthropic, maker of Claude—which had, till not too long ago, been the one AI mannequin cleared for Pentagon categorised networks.
What precisely did these seven corporations comply with
The agreements cowl “lawful operational use”—suppose battlefield choice help, knowledge synthesis, situational consciousness, and intelligence work. Every firm’s AI will probably be built-in into the Pentagon’s categorised IL6 and IL7 environments.The contracts embrace language stating AI shouldn’t be used for “home mass surveillance or autonomous weapons with out applicable human oversight.” However there is a catch: not one of the corporations get a veto over how the Pentagon in the end deploys the expertise. That distinction is strictly what blew up negotiations with Anthropic.The Pentagon’s GenAI.mil platform, working since December with Google’s Gemini mannequin, has already seen over 1.3 million division personnel use it—producing tens of tens of millions of prompts and a whole lot of 1000’s of AI brokers in simply 5 months.
Google workers despatched a letter however Google signed anyway
The timing right here is notable. Simply days earlier than the Pentagon’s announcement, round 600 Google workers despatched an open letter to CEO Sundar Pichai urging him to refuse categorised navy work fully.“We need to see AI profit humanity; to not see it being utilized in inhumane or extraordinarily dangerous methods,” the workers wrote, flagging deadly autonomous weapons and mass surveillance as particular issues. “Making the mistaken name proper now would trigger irreparable harm to Google’s popularity, enterprise, and position on this planet.”Google signed the deal anyway. An organization spokesperson mentioned it “stays dedicated to the consensus that AI shouldn’t be used for home mass surveillance or autonomous weaponry with out applicable human oversight”—however that offering API entry to its fashions represents “a accountable method to supporting nationwide safety.”It is a replay of 2018, when Google pulled out of Undertaking Maven after worker backlash. This time, management did not blink.
Why is not Anthropic within the room
Anthropic drew two agency traces: no totally autonomous weapons, no mass home surveillance. The Pentagon needed blanket “any lawful use” language. Talks collapsed in February.The Warfare Division then declared Anthropic a “provide chain danger”—a designation traditionally used for overseas adversaries, by no means earlier than utilized to a US firm. Anthropic sued. A federal choose blocked the designation in March, although an appeals courtroom later declined to completely carry it. The authorized battle is ongoing.Protection Division CTO Emil Michael confirmed Friday that Anthropic stays blacklisted—although he individually flagged the corporate’s highly effective new cybersecurity mannequin, Mythos, as a “separate nationwide safety second” that your entire authorities wants to concentrate to. The NSA is reportedly already utilizing it, designation however.
A take care of Anthropic would possibly nonetheless occur, simply not but
Behind the scenes, the White Home seems to be on the lookout for an exit ramp. A draft government motion is reportedly within the works that would give companies a path to work with Anthropic once more—and particularly to entry Mythos, which has rattled each authorities officers and banks with its skill to seek out vulnerabilities in well-tested software program.President Trump instructed CNBC it is “potential” a deal might occur, calling Anthropic “very good” and doubtlessly of “nice use.” Anthropic CEO Dario Amodei met with senior White Home officers earlier this month, with either side describing the dialog as productive.The announcement, although, sends a transparent message: the Pentagon is not ready. It is diversifying its AI stack, locking in rivals, and making Anthropic’s absence more and more costly. With vital federal funding earmarked for AI and offensive cyber operations below the One Massive Stunning Invoice Act, the businesses now on the desk have a big head begin.