Pentagon's AI push faces caution from military leaders over battlefield use

The Trump administration's push to rapidly expand the use of artificial intelligence (AI) across the US military is facing calls for caution from senior military leaders and technology companies concerned about the risks associated with deploying the technology on the battlefield.

Adm. Frank Bradley, head of the US Special Operations Command (SOCOM), recently warned that armed forces must proceed carefully when integrating AI into lethal military operations.

Speaking at a special forces conference in Tampa, Florida, Bradley said troops “have to be very careful” about how AI is employed in delivering lethal force. While he envisaged a future in which AI could help identify military targets, he stressed that human operators must remain confident that force would be applied only where intended.

The remarks highlight a growing debate within the US military as Defence Secretary Pete Hegseth pushes for wider adoption of AI technologies. Hegseth has argued that the Pentagon should be free to use AI in any lawful military application and has opposed restrictions that could limit battlefield use of the technology.

President Donald Trump has also emphasised maintaining America's lead in AI development, recently shelving plans for a new executive order after expressing concerns that additional regulations could weaken the country's competitive advantage over rivals such as China.

Within the military, however, views on AI's role vary.

A Pentagon official said current efforts are focused on developing battlefield tools that can help troops identify and process targets more quickly, accelerating military decision-making and strike operations. At the same time, SOCOM officials described AI primarily as a support tool designed to reduce administrative burdens and improve efficiency rather than replace human judgement.

SOCOM's senior enlisted leader Sgt. Maj. Andrew Krogman said AI could help automate routine tasks and modernise military processes, while acquisition official Melissa Johnson said the technology should reduce the cognitive burden associated with mundane work.

“We're leveraging AI more and more, but it's not to replace operator judgment, it's to enhance it,” Johnson said.

Military officials have already acknowledged operational uses of AI. Lt. Gen. Michael Conley, head of Air Force Special Operations Command, told lawmakers earlier this month that AI tools were used during the Iran conflict to rapidly downgrade highly classified intelligence, allowing it to be shared more quickly with drone operators in the field.

Experts say AI is also increasingly being used to support military targeting. Helen Toner of Georgetown University's Center for Security and Emerging Technology said AI is enabling armed forces to operate at greater speed and scale while humans continue to make key decisions.

Pentagon-Anthropic dispute

The debate over military AI has also triggered a public dispute between the Pentagon and artificial intelligence company Anthropic.

The company has expressed concerns over unrestricted government use of AI, including fully autonomous weapons and AI-assisted surveillance systems. After Anthropic chief executive Dario Amodei raised objections regarding the use of its Claude chatbot within classified military networks, the Pentagon terminated a $200 million defence contract and designated the company a supply-chain risk.

Anthropic has since filed a lawsuit, alleging that the Pentagon's actions constitute unlawful retaliation. The Defence Department has instead highlighted partnerships with other technology firms, including Google, OpenAI and SpaceX, to develop AI systems aimed at enhancing military decision-making in complex operational environments.

Despite concerns, experts note that military organisations traditionally approach new technologies cautiously because of the potentially severe consequences of mistakes.

“Commanders want their missions to succeed, which means both being able to create lethal effects at scale, and avoiding unintended effects like friendly fire, civilian casualties, or simply identifying targets incorrectly,” Toner said.