US Military to Use AI Chatbots for Ranking Targets, Official Reveals
Key Facts
- What: A Defense Department official disclosed that generative AI chatbots could analyze target lists, rank them by priority based on factors like aircraft location, with final decisions remaining with humans.
- Models: OpenAI’s ChatGPT and xAI’s Grok recently gained approval for classified Pentagon use; Anthropic’s Claude has already been integrated into operations in Iran and Venezuela.
- Context: The disclosure comes amid scrutiny over a US strike on an Iranian girls’ school that killed more than 100 children, partly blamed on outdated targeting data.
- Technology Split: Traditional Maven AI (computer vision, since 2017) handles imagery analysis; generative AI adds a conversational layer for faster prioritization.
- Timeline: OpenAI agreement signed February 28; Anthropic designated a supply chain risk after disputes, with Trump calling for phase-out within six months.
The US military is planning to deploy generative AI chatbots to help rank potential targets and recommend strike priorities, a senior Defense Department official has revealed. The systems would ingest lists of possible targets, analyze them in natural language, and suggest an order of operations while factoring in real-time variables such as aircraft positioning. Humans would remain responsible for vetting every recommendation.
The comments, shared on background with MIT Technology Review, provide the clearest picture yet of how the Pentagon intends to integrate large language models into life-and-death targeting decisions. They arrive as the military faces intense scrutiny following a deadly strike on a girls’ school in Iran that killed over 100 children, an incident still under investigation and partly attributed to outdated targeting data.
The official described the chatbot approach as a possible future workflow rather than a confirmed current practice. However, the disclosure underscores the Pentagon’s rapid acceleration of AI across classified networks, even as questions mount about reliability, accountability, and the speed at which human oversight can truly function.
How the Chatbot Targeting Process Would Work
According to the official, a list of potential targets would be fed into a generative AI system approved for classified environments. Military personnel could then query the model in natural language, asking it to rank targets by priority while considering constraints such as current aircraft locations, available munitions, or other operational factors.
This represents a significant evolution from the Pentagon’s existing “Project Maven” system, launched in 2017. Maven primarily relies on older computer vision AI to scan massive volumes of drone footage and imagery, automatically flagging potential targets. A 2024 Georgetown University report documented soldiers using Maven’s map-based dashboard to select and vet targets, dramatically shortening approval times.
The new generative AI layer would function as a conversational interface on top of that data. Instead of solely interpreting highlighted dots on a battlefield map, operators could simply ask the chatbot for prioritized recommendations. The official noted this approach accelerates the search for targets and the overall targeting process, though no specific time savings were provided given the requirement for human double-checking.
Contrasting AI Technologies and Their Risks
The Pentagon is now deploying two fundamentally different forms of AI with distinct limitations. Maven’s computer vision models are more “battle-tested” and tied to direct visual data inspection. Generative AI systems like ChatGPT, Claude, and Grok, built on large language models, are far newer to military applications and harder to verify.
“Generative AI systems... are a fundamentally different technology from the AI that has primarily powered Maven,” the MIT Technology Review report notes. While easier to interact with through natural language, their outputs can be more opaque and prone to hallucination — a serious concern when the stakes involve human lives.
Anthropic’s Claude has reportedly already been used in operations in Iran and Venezuela, including efforts that led to the capture of Venezuelan leader Nicolas Maduro in January. However, tensions between the Pentagon and Anthropic escalated after the company sought to restrict certain military uses of its technology. The Defense Department labeled Anthropic a supply chain risk, and President Trump publicly demanded the government stop using its AI products within six months. Anthropic is challenging the designation in court.
In the wake of that dispute, OpenAI secured an agreement on February 28 allowing its technology to be used in classified settings. xAI has also reached a similar deal, positioning both companies to potentially supply the chatbots used in future targeting workflows.
Increased Scrutiny After Deadly Iran Strike
The timing of the disclosure has heightened concerns. Multiple news outlets have linked the US military to a strike on an Iranian girls’ school in which more than 100 children died. The Washington Post has reported that both Claude and Maven were involved in targeting decisions in Iran, though the precise role of generative AI remains unclear.
The New York Times reported that a preliminary investigation found outdated targeting data partly responsible for the tragedy. The Pentagon continues to investigate and has not confirmed the use of any specific AI system in that strike.
Beyond classified operations, the military has rolled out non-classified generative AI tools to millions of service members through GenAI.mil since December for tasks like contract analysis and presentation writing. However, only a handful of models have received approval for the more sensitive classified networks where targeting occurs.
Impact
The integration of AI chatbots into targeting decisions could dramatically change the pace of modern warfare. By allowing operators to query systems conversationally, the military hopes to synthesize vast amounts of intelligence faster than ever before.
“The use of generative AI for such decisions is reducing the time required in the targeting process,” the Defense official told MIT Technology Review.
This speed comes with trade-offs. While human oversight remains mandatory, the ease of generating AI recommendations may create pressure to move more quickly, potentially reducing the thoroughness of verification. Critics worry that opaque “black box” outputs from large language models could make it harder for operators to spot errors compared to the more transparent visual dashboards of systems like Maven.
For developers and AI companies, the Pentagon’s embrace represents a massive new market. OpenAI and xAI have gained ground after Anthropic’s difficulties, highlighting how political and contractual battles can rapidly reshape which AI models power military operations. The deals also raise profound ethical questions about the commercialization of AI for lethal decisions.
What's Next
The Pentagon continues pushing AI companies to expand capabilities on classified networks. Reuters has reported that military officials are actively seeking broader deployment of these tools for mission planning and weapons targeting.
Anthropic’s legal battle over its supply chain risk designation will likely determine whether its models remain available. Meanwhile, OpenAI and xAI are positioned to fill the gap, though the reliability of their systems in high-stakes combat environments remains unproven publicly.
The military’s broader AI ambitions suggest chatbots for targeting prioritization are only the beginning. As more models gain classified approval, the line between intelligence analysis and automated recommendation will continue to blur, raising urgent questions about accountability when AI-assisted strikes go wrong.
The official’s comments provide a rare window into the Pentagon’s thinking, demonstrating that generative AI is moving from support roles into core operational decision-making. How effectively humans can oversee these powerful but imperfect systems may determine whether this technology saves lives by improving precision or risks them through accelerated, harder-to-verify decisions.
Sources
- Defense official reveals how AI chatbots could be used for targeting decisions | MIT Technology Review
- OpenAI Reaches A.I. Agreement With Defense Dept. After Anthropic Clash - The New York Times
- Exclusive: Pentagon pushing AI companies to expand on classified networks, sources say | Reuters
- AI on the battlefield: How is the US integrating AI into its military? | Euronews