San Francisco: US artificial intelligence firm Anthropic has refused to grant the US Department of Defence unrestricted access to its AI systems, setting up a high-stakes standoff over military use of advanced technology.

The company said it would continue supplying tools for defensive and intelligence purposes but drew a red line against applications such as mass domestic surveillance and fully autonomous weapons.

"These threats do not change our position: we cannot in good conscience accede to their request," chief executive Dario Amodei said.

Also read | Workplace AI gains urgency as 91 pc Indian firms test solutions

Washington had reportedly given Anthropic until Friday evening to agree to unconditional military use of its technology, even in ways that could conflict with the firm’s ethical guidelines, or risk being compelled under emergency federal powers.

Amodei said Anthropic’s models are already used by the Pentagon and intelligence agencies for national defence, but warned that certain uses would cross fundamental democratic and safety boundaries.

"Using these systems for mass domestic surveillance is incompatible with democratic values," he said.

Also read | AI video detection tools: Do they really work or are they flawed?

He also cautioned that current AI systems are not reliable enough to operate lethal weapons without human oversight.

"We will not knowingly provide a product that puts America's warfighters and civilians at risk."

Pentagon ultimatum

Following meetings earlier this week, the Pentagon invoked the possibility of using the Defence Production Act, a Cold War-era statute that allows the government to compel private companies to prioritise national security needs. The law was last widely used during the COVID-19 pandemic.

Officials also warned that Anthropic could be designated a “supply chain risk”, a label typically applied to firms from adversary nations, a move that could severely restrict its ability to win US government contracts.

Also read | ‘Claude AI killed my startup’: Founder says Anthropic update wiped out her AI business overnight

A senior Pentagon official rejected the company’s concerns, saying the department operates strictly within legal limits.

"Legality is the Pentagon's responsibility as the end user," the official said, adding that the department "has only given out lawful orders."

The official confirmed that discussions between Anthropic and the Pentagon had included scenarios involving intercontinental ballistic missiles, underscoring the sensitivity of the dispute.

Competitive pressure

The Pentagon said Grok, developed by Elon Musk’s xAI, has already been cleared for use in classified environments. Other contractors, including OpenAI and Google, are said to be close to receiving similar approvals, increasing pressure on Anthropic to comply.

Anthropic is part of a $200 million Pentagon programme to supply AI models for military applications.

Safety vs national security

Founded in 2021 by former OpenAI researchers, Anthropic has positioned itself around a safety-first approach to AI development, a stance now putting it at odds with US defence officials.

"Anthropic understands that the Department of War, not private companies, makes military decisions," Amodei said.

"However, in a narrow set of cases, we believe AI can undermine, rather than defend, democratic values."

What the controversy is about

The dispute centres on whether private AI companies should be compelled to provide unrestricted access to their most powerful systems for military use. The Pentagon argues that national security needs must take priority, while Anthropic insists certain applications, particularly domestic surveillance and autonomous lethal weapons, violate its ethical standards and could pose risks to civilians and democratic norms.