The artificial intelligence company OpenAI has quietly joined a high-stakes U.S. military initiative exploring how voice commands could direct swarms of autonomous drones, according to people familiar with the effort. The collaboration is part of a $100 million Pentagon prize challenge aimed at rapidly developing software capable of coordinating large numbers of unmanned systems in combat scenarios.
The competition, launched jointly by the U.S. Department of Defense, Special Operations Command, and the Defense Innovation Unit, seeks prototypes that can translate battlefield instructions into synchronized drone action across air and sea. The six-month challenge is structured in phases, allowing only the most promising teams to advance to live-platform demonstrations.
According to individuals involved in the submissions, the AI firm’s role is limited to converting spoken orders from commanders into digital instructions, not piloting the drones, selecting targets, or integrating weapons. Only an open-source version of its models would be used, and the company has not submitted its own bid. Instead, two partner organizations incorporated its technology into their proposals.
One of the qualifying teams is led by Applied Intuition, working alongside Sierra Nevada Corporation and Noda AI. Internal documents describe a system in which human speech is translated into operational commands within a software layer positioned between operators and autonomous machines, while other partners handle integration and swarm coordination.
Defense officials have made clear that the program is designed with combat effectiveness in mind, stating that improved human-machine interaction will directly influence the “lethality and effectiveness” of such systems. Example commands could include directing entire groups of unmanned vessels to reposition simultaneously, illustrating the scale of automation envisioned.
The involvement of a leading generative-AI developer marks a notable expansion of Silicon Valley’s relationship with the military. Earlier this week, the Pentagon also announced a separate arrangement to make the company’s chatbot tools available to millions of Defense Department personnel for administrative and analytical use.
Chief Executive Sam Altman has previously expressed caution about weaponized AI, saying in 2024 that most of the world does not want artificial intelligence making lethal decisions, though he stopped short of ruling out future involvement entirely. His comments reflect an ongoing tension between commercial AI ambitions and ethical concerns about military deployment.
Some defense officials and technologists remain uneasy about integrating generative AI into weapons-adjacent systems, citing risks such as bias, errors, or so-called hallucinations in large language models. Several people familiar with the project emphasized the importance of keeping humans firmly “in the loop” to prevent automated misinterpretation of battlefield instructions.
The debate echoes earlier controversies over military tech partnerships, including employee protests at Google in 2018 tied to the Pentagon’s Project Maven. More recently, however, AI companies have shown greater willingness to engage in national security work, with policy revisions and new collaborations such as OpenAI’s tie-up with Anduril Industries focused on counter-drone defense. Industry peer Anthropic has also faced internal debate over the societal implications of advanced AI systems.
Technologically, commanding large autonomous swarms remains a formidable challenge. While militaries can already operate multiple drones simultaneously, coordinating them as adaptive, semi-independent units capable of executing complex missions is still an emerging capability. Early phases of the Pentagon challenge will focus on software translation and coordination, with later stages expected to address targeting awareness and full mission execution.
As global powers race to integrate artificial intelligence into defense strategies, the experiment underscores a broader shift: AI is moving from back-office analytics toward direct operational relevance on the battlefield, even as governments and developers struggle to define the limits of its use.
