During the Pentagon dispute, Anthropic entered a drone swarm competition

Anthropic was among several AI firms that submitted a proposal earlier this year to compete in a $100 million Pentagon prize challenge aimed at developing voice-controlled, autonomous drone swarming technology, according to people familiar with the matter.

The submission came during tense negotiations between Anthropic and the US Defense Department over the limits, or “red lines,” on how its AI systems could be used by the military. On Friday, Defense Secretary Pete Hegseth ordered the Pentagon to bar contractors and their partners from engaging in commercial activity with Anthropic, escalating tensions.

At the centre of the dispute is the potential role of AI in so-called “killer robots” — weapons systems capable of selecting and firing at targets without direct human involvement. Anthropic executives have repeatedly said they support extensive lawful military uses of AI, short of mass domestic surveillance and fully autonomous weapons.

According to one of the people familiar with the proposal, Anthropic believed its submission did not cross those red lines. While the broader programme could eventually enable lethal drone swarms, a human operator would still be able to monitor and intervene if necessary. The company’s proposal focused on using its Claude AI tool to translate a commander’s voice instructions into digital commands and to coordinate fleets of drones across domains such as air and sea. It did not involve autonomous targeting or weapons decisions, and human oversight would remain in place.

Anthropic was not ultimately selected for the competition, though the reason remains unclear. The company declined to comment, maintaining its broader position that AI systems are not yet sufficiently reliable to independently operate autonomous weapons.

The prize challenge was jointly launched by US Special Operations Command, which oversees the Defense Autonomous Warfare Group, and the Defense Innovation Unit. The initiative is structured in five phases, beginning with software development before moving to live testing on physical platforms. A Pentagon official previously indicated that the systems are intended for offensive use, stating that human-machine interaction would directly affect their “lethality and effectiveness.” Later stages of the programme include “target-related awareness and sharing” and ultimately “launch to termination” capabilities.

Among the successful bidders were proposals from SpaceX and xAI. Two defence technology companies that partnered with OpenAI were also selected, including a submission led by Applied Intuition. In that bid, OpenAI’s technology would be used in mission control systems to translate voice commands into digital instructions. An OpenAI spokesperson said any support provided would remain within the company’s usage policies.

Hours after the Pentagon sidelined Anthropic, OpenAI announced a new agreement with the Defense Department to deploy its AI tools on classified cloud systems. Chief executive Sam Altman said the deal would require “human responsibility for the use of force, including for autonomous weapon systems.”

While multiple drones can already be operated simultaneously, experts say building reliable software capable of directing large swarms autonomously across air and sea remains technically complex. The Pentagon’s competition will proceed in phases, depending on participant performance and continued interest.