Pentagon AI chief Emil Michael says Anthropic is – Business News
Pentagon AI chief Emil Michael isn’t sweating Microsoft’s legal assist for Anthropic. In reality, he advised me the entire scenario is “totally bananas.”
Michael dismissed Microsoft’s amicus temporary, which the company filed earlier this week in assist of Anthropic and towards the Department of War, as little more than a transfer to appease staff.
“It’s sort of par for the course to prove to your employees that you have certain camaraderie [with other left-wing tech companies]”, mentioned Michael, who serves as Under Secretary of Defense for Research & Engineering and oversees AI contracts, on a telephone call earlier this week. “If you read their actual briefs, it says everything the Department of War wants to do is legal.”
Emil Michael (proper, with Peter Hegseth) serves as Under Secretary of Defense for Research & Engineering and oversees AI contracts. Getty Images
While the amicus temporary claims Microsoft and Anthropic need to shield warfighters, Michael isn’t shopping for it.
It’s the federal government, he advised me, that’s working to keep service members secure.
“[Our move designating Anthropic a supply chain risk] is to protect warfighters — because Anthropic was trying to insert themselves into the chain of command,” Michael defined. “Supply chain risk is to remove that risk, not add another risk.”
Michael mentioned this is the newest wrinkle in a saga made more weird by the truth that Anthropic labored carefully with the Department of War for years earlier than the federal government terminated its contract in February.
Following the January 2026 US raid that captured and extradited Nicolás Maduro from Venezuela, an Anthropic government requested a counterpart at Palantir — through which Anthropic’s Claude AI was built-in — whether or not and how its artificial intelligence was used.
Anthropic (CEO Dario Amodei is pictured right here) labored carefully with the Department of War for years earlier than abruptly drawing the road. REUTERS
Anthropic had provided a specialised model of Claude to some of the army’s most delicate instructions — together with Central Command and Indo-Pacific Command — underneath a deal price roughly $200 million.
Palantir advised Pentagon officers about Anthropic’s inquiry, and the Feds interpreted it as potential post-facto disapproval. The transfer raised alarms over dependency dangers on an AI supplier which may second-guess categorized operations, triggering a contract review and failed talks.
Anthropic insisted on safeguards barring mass home surveillance of US residents — which the US denies doing — and totally autonomous deadly weapons with out human oversight. In response, the Pentagon labeled Anthropic a “supply chain risk” on March 4 and changed Anthropic’s technology with OpenAI.
Anthropic sued on March 9, difficult it as illegal and retaliatory.
The Pentagon’s concern is that its contractors may attempt to dictate their operations. “I don’t want and can’t have a model that has a policy bias that could be poisoned because it doesn’t want to do certain military activities — or an insider threat who disagrees with the democratically elected leadership of this country,” Michael mentioned.
It’s a stance Michael finds onerous to sq. with Anthropic’s own claims about its technology.
“They’re saying this technology is so powerful — more powerful than a nuclear weapon, more powerful than some governments — [and] they don’t want to use that power to help our country or our war department,” Michael advised me in disbelief.
“This would never happen in China … And yet they don’t want to serve the most important part of the government in time of conflict. That seems very off.”
This story is half of NYNext, an indispensable insider insight into the improvements, moonshots and political chess strikes that matter most to NYC’s energy gamers (and people who aspire to be).
He mentioned what makes it stranger is that this comes after years of working with the Pentagon.
“What’s very surprising is for a company that didn’t want to do Department of War stuff, why are they selling and deploying their software to the Department of War for three years?” he mentioned.
The episode was a reminder that regardless of corporations like Palantir or Anduril enthusiastically supporting America and taking contracts supporting Customs and Border Patrol and ICE, a slice of Silicon Valley nonetheless resists supporting the army.
The Pentagon labeled Anthropic a “supply chain risk” — a designation usually reserved for international adversaries — and OpenAI stepped in to take the contract. AP
“When I took this job, I thought those 2018 days where Google’s employees revolted were behind us,” he mentioned, referencing the worker rebellion that compelled Google to desert Project Maven, a Pentagon contract to make use of AI for drone imagery evaluation. The incident has grow to be a defining image of Silicon Valley’s resistance to army work.
But today, he believes Anthropic is the exception, not the norm. “I don’t think Elon’s woke with AI and Grok,” he mentioned. And even a company like Google? They “learned the lesson of 2018.”
As the Pentagon irons out huge contracts with longtime companions, it’s additionally investing in new suppliers to make sure the US is all the time forward.
In the approaching weeks, Michael mentioned, the division will announce a wave of new partnerships with smaller, non-traditional protection corporations — startups unburdened by the ideological baggage that derailed Anthropic. For Michael, it’s the proper ending to a unusual saga: the long run of American army AI, constructed by corporations that truly need the work.
He mentioned, “[They’re] building stuff for the next generation of warfare.”
