Anthropic, one of the world’s most closely watched AI companies, had put forward a proposal earlier this year for a US Pentagon competition aimed at building voice-controlled drone swarm technology. The submission, according to people aware of the development, was part of a $100 million prize challenge designed to accelerate research into autonomous military systems that can operate across air and sea.
The move is notable because it came at a time when Anthropic was in delicate discussions with the US Department of Defense over the limits it wanted to place on how its AI models could be deployed in military settings. The company has consistently maintained that while it supports lawful military applications of AI, it does not back the development of fully autonomous weapons that can independently choose and attack targets, nor does it endorse mass domestic surveillance.
Anthropic wanted to use Claude AI to coordinate US drone swarms in Pentagon challenge
People familiar with the proposal told Bloomberg that Anthropic did not view its entry into the contest as crossing its own internal boundaries. The submission reportedly centred on using its Claude AI model to interpret a commander’s spoken intent and convert it into structured digital commands. These instructions would then be used to coordinate a fleet of drones operating together as a swarm.
Crucially, the AI system was not meant to take decisions related to targeting or weapons release. Instead, human operators would retain supervision over the system and would have the ability to intervene or halt operations if necessary. In other words, the company’s role would have focused on enabling coordination and communication, rather than delegating lethal decision-making to software.
The Pentagon’s competition is structured as a phased programme. It begins with software development before moving towards real-world testing on live platforms. According to a January announcement by defence officials, the initiative is intended for offensive capabilities, with the human-machine interface expected to influence “the lethality and effectiveness of these systems.” The early stage concentrates on building software capable of synchronising drone movements across multiple domains, including air and maritime environments. Later stages envision features such as target awareness, information sharing and, eventually, systems that can operate from “launch to termination.”
Human oversight at the centre of AI defence debate in the US
Despite its participation, Anthropic was not selected in the initial round of winners. The reasons for this have not been made public.
The context became more complicated on Friday when US Defense Secretary Pete Hegseth directed the Pentagon to prevent contractors and their partners from engaging in commercial dealings with Anthropic. The order followed tensions over how AI companies should engage with defence projects involving autonomous capabilities.
Anthropic has previously argued that AI systems are not yet reliable enough to independently run autonomous weapons platforms. At the same time, company executives have said they are open to supporting military use cases that comply with international law and preserve meaningful human control.
The drone swarm challenge was launched jointly by US Special Operations Command, which oversees the Defense Autonomous Warfare Group, and the Defense Innovation Unit. The programme is expected to proceed in stages, depending on participant progress and continued interest.
Other technology players fared better in the selection process. A proposal involving SpaceX and xAI was among those chosen. Two defence technology firms that listed OpenAI as an AI partner were also selected. One of these bids was led by Applied Intuition, known for its work on autonomous vehicle systems.
In OpenAI’s case, its tools would reportedly assist in mission control functions by converting voice inputs into digital operational commands. A spokesperson for the company said that any involvement would remain within its established usage policies.
Soon after the Pentagon’s move concerning Anthropic, OpenAI chief executive Sam Altman announced a fresh agreement with the Defense Department to deploy OpenAI’s systems on classified cloud infrastructure. According to Altman, the arrangement would require “human responsibility for the use of force, including for autonomous weapon systems.”
While armed forces around the world already deploy multiple drones simultaneously, orchestrating large numbers of them into an intelligent, coordinated swarm remains technically demanding. Such systems must manage communication, navigation and adaptability in dynamic combat environments. The Pentagon’s challenge underscores how rapidly defence agencies are moving to harness AI for these complex tasks, even as companies like Anthropic attempt to draw ethical lines around their participation.
