• OpenAI CEO Sam Altman informed employees the company is negotiating a potential solution to the standoff between Anthropic and the Pentagon over unrestricted military AI access.
  • The Pentagon has imposed a Friday evening deadline for Anthropic to grant "all lawful use" of its Claude AI or face business termination and possible Defense Production Act enforcement.
  • Employee backlash is mounting, with over 220 verified signers from Google and OpenAI petitioning against military AI for surveillance or autonomous killing without oversight.

OpenAI CEO Sam Altman told staff in an internal meeting Thursday that the company is working on a deal that could help resolve the escalating impasse between AI startup Anthropic and the Pentagon, according to people familiar with the matter. The development comes as tensions reach a critical point, with the Pentagon threatening to invoke the Defense Production Act if Anthropic doesn't comply with demands for full access to its Claude large language model by Friday evening.

Anthropic CEO Dario Amodei has rejected Pentagon demands for safeguards against Claude's use in mass surveillance of Americans or fully autonomous weapons, calling them non-negotiable guardrails. This stance has put the company's $200 million Pentagon contract from July 2025 at immediate risk. "Without a deal, Anthropic would be forced into offboarding from defense work," said one source close to the negotiations, who spoke on condition of anonymity.

Defense Secretary Pete Hegseth met with Amodei on Tuesday, demanding a signed document granting "all lawful use" of Claude for national security purposes. Pentagon officials have labeled Anthropic's restrictions as creating "supply chain risk" in what Hegseth has described as a "wartime arms race" requiring rapid AI deployment. Efforts to restructure the agreement have hit a snag despite what one official called "superficial concessions" in the Pentagon's "best and final" offer Wednesday.

Meanwhile, employee resistance is coalescing across the industry. More than 220 verified signers—including 176 from Google and 47 from OpenAI—have joined a "We Will Not Be Divided" petition explicitly opposing the Pentagon's moves on Anthropic. The petition urges unified refusal against military AI for surveillance or autonomous killing without proper oversight, framing the situation as Pentagon "divide and conquer" tactics. "We're seeing unprecedented solidarity against what feels like quasi-nationalization of AI capabilities," said a Google employee who signed the petition but requested anonymity due to fear of retaliation.

OpenAI's potential involvement as an intermediary marks a significant development, though details of Altman's proposed solution remain unclear. The company has faced its own negotiations with the Pentagon over AI access, creating complex dynamics. When reached for comment, an OpenAI spokesperson declined to elaborate beyond confirming that "discussions are ongoing to support responsible AI deployment for national security."

Historical precedent looms large—Google employee protests led to the company's withdrawal from Project Maven in 2018, while current tensions mark what former White House AI advisor Dean Ball calls "uncharted territory" for potential DPA use against an AI firm. Ball warned of "huge risk" and signals that government dealings are "extremely dangerous" for tech companies.

If Anthropic rejects the Pentagon's ultimatum, officials plan to pivot immediately to OpenAI and Google while considering DPA enforcement. This could trigger what one defense analyst described as "business ruin" for Anthropic or broader sector-wide pressure. Hegseth has stated there's "no tolerance for restricted models in war," suggesting accelerated military AI push regardless of outcome.

Correction: An earlier version misstated the number of petition signers from OpenAI; the correct figure is 47.