How talks between anthropic and the Defense Dept. fell apart.
- The San Juan Daily Star
- 7 hours ago
- 5 min read
By SHEERA FRENKEL, CADE METZ and JULIAN E. BARNES
Minutes before a 5:01 p.m. deadline Friday, Emil Michael, the Defense Department’s chief technology officer, was fuming.
For weeks, Michael, a former top executive at Uber, had been negotiating a $200 million artificial intelligence contract with the AI company Anthropic for the Pentagon. The talks had hit obstacles as the agency demanded unfettered use of Anthropic’s AI systems, while the company countered that it would not allow its technology to be used for purposes such as the surveillance of Americans.
Defense Secretary Pete Hegseth had set the Friday deadline for a deal and the two sides were close. The only thing that remained was agreeing on a few words about the issue of lawful surveillance of Americans, multiple people with knowledge of the talks said.
Michael, who was on a call with Anthropic executives, demanded that the company’s CEO, Dario Amodei, get on the phone to hash out the language, the people said. But Michael was told that Amodei was in a meeting with his executive team and needed more time.
Michael was unhappy with that answer, the people said. He also had an ace up his sleeve: On the side, he had been hammering out an alternative to Anthropic with its rival, OpenAI. A framework between the Pentagon and OpenAI had already been reached.
So when the Friday deadline passed, the Defense Department did not give Anthropic more time. At 5:14 p.m., Hegseth announced that he had designated Anthropic as a security risk and that it would be cut off from working with the U.S. government. “America’s warfighters will never be held hostage by the ideological whims of Big Tech,” he posted on social media.
Later that night, Sam Altman, OpenAI’s CEO, announced that his company had instead reached an agreement with the Pentagon to provide its AI technologies for classified systems.
In the end, the talks between Anthropic and the Defense Department were undone by weeks of building frustration between men who had differing philosophies about AI and who did not like one another.
This account of the failure of the Anthropic talks and the success of the OpenAI deal is based on interviews with a dozen people with knowledge of the negotiations. The New York Times spoke to people from multiple companies and government agencies and interviewed officials with a wide range of views on the fight over the future of AI in warfare.
Michael, Amodei and Altman have known one another for years through business dealings in Silicon Valley, but they have often not gotten along. Amodei and Altman, 40, once worked together at OpenAI and are bitter rivals. And as Anthropic’s discussions with the Defense Department dragged on last week, Michael, 53, publicly accused Amodei of being “a liar” with “a God-complex.”
Ultimately, Michael preferred Altman — who has courted the Trump administration — over Amodei, the people with knowledge of the negotiations said.
The clashes between the Defense Department and Anthropic are most likely not over. On Friday, Anthropic said it would sue over the Pentagon’s decision to label it a “supply chain risk.” The supply chain risk designation has typically been reserved for foreign companies that the U.S. government believes are a threat to national security; the label has never been used against an American company.
Officials at U.S. intelligence agencies including the CIA, which uses Anthropic’s AI technology, have also privately urged both sides to make a deal. Some current and former officials said they continued to hope for a peace agreement.
(The New York Times has sued OpenAI and Microsoft, accusing them of copyright infringement of news content related to AI systems. The companies have denied those claims.)
Last year, Anthropic, OpenAI, Google and xAI were all part of a Pentagon pilot program to explore how AI could be used for defense. Anthropic was the only AI company that deployed its technologies to work on classified systems and its AI was widely used by defense officials.
Michael had joined the Defense Department as chief technology officer in May 2025 after previously working as a special assistant at the Pentagon during the Obama administration. Michael became the point person on the negotiations with Anthropic.
But the talks soon reached an impasse. Anthropic wanted guardrails to stop its AI from being used for the mass surveillance of Americans or deployed in autonomous weapons with no humans involved. The Defense Department argued that no private contractor could decide how its tools would be lawfully used.
On Feb. 24, Hegseth called a meeting with Amodei at the Pentagon to find a resolution. The men showed little warmth in the meeting, which lasted less than an hour, people familiar with the discussions said.
At the end of the conversation, Hegseth said that if Anthropic did not compromise with the Pentagon by 5:01 p.m. Friday, it would be labeled a supply chain risk. He said the Pentagon could also invoke the Defense Production Act to force Anthropic to work with the government, a move that was later dropped.
The next day, Altman of OpenAI got on a call with Michael to discuss a deal for his company. Within a day, they had drafted a rough framework. OpenAI agreed to the Pentagon’s requirement that its AI could be used for all lawful purposes, but it also negotiated the right to put technical guardrails on its systems to adhere to its safety principles.
Amodei doubled down on AI safety. In a statement on Feb. 26, he said Anthropic could not “in good conscience accede” to the Pentagon’s demands.
“In a narrow set of cases, we believe AI can undermine, rather than defend, democratic values,” he added. “Some uses are also simply outside the bounds of what today’s technology can safely and reliably do.”
That night, Michael unleashed on Amodei on social media, calling the Anthropic leader a liar. “He wants nothing more than to try to personally control the US Military and is ok putting our nation’s safety at risk,” Michael posted.
As Friday’s deadline approached, Anthropic executives thought they were close to a compromise with the Pentagon and were just a few words apart on the issue of surveillance, people on both sides of the negotiation said.
Complicating the matter was a social media post by President Donald Trump. Trump had told Hegseth on Friday morning that he had prepared a post belittling Anthropic and ordering all government agencies to stop working with it within six months.
Even as Trump published the post at 3:47 p.m., the two sides kept talking. Michael, who was on a call with Anthropic executives at the time, said the Pentagon wanted the company to allow for the collection and analysis of unclassified, commercial bulk data on Americans, such as geolocation and web browsing data, people briefed on the negotiations said.
Anthropic told the Pentagon that it was willing to let its technology be used by the National Security Agency for classified material collected under the Foreign Intelligence Surveillance Act. But the company wanted a legally binding promise from the Pentagon not to use its technology on unclassified commercial data.
At that point, Michael asked to speak with Amodei, who was not on the call. Michael was told that Amodei was in a meeting. Shortly after, Hegseth said the talks were over.
At 10 p.m. Friday, as Anthropic’s lawyers began working on a lawsuit against the Pentagon, Altman was on the phone with Michael finalizing the details of OpenAI’s deal with the Defense Department. Altman then posted news of the agreement on social media. Hegseth later reposted Altman’s announcement from his personal account on the social platform X.
Comments