Pentagon Eyes AI Chatbots for Military Targeting Decisions

When a Defense Department official sat down with MIT Technology Review this week, they revealed something that would have seemed like science fiction just a few years ago: the Pentagon is actively considering using generative AI chatbots to help rank military targets and recommend which ones to strike first.

“The US military might use generative AI systems to rank lists of targets and make recommendations—which would be vetted by humans—about which to strike first.” — Defense Department Official

The New Battlefield: Generative AI Meets Military Operations

The disclosure comes at a critical moment. The Pentagon is currently investigating a strike on an Iranian girls’ school that killed more than 100 children, an incident that has intensified scrutiny over how AI systems are being integrated into military decision-making.

According to the official, who spoke on background to discuss sensitive matters, the process would work like this: a list of possible targets gets fed into a generative AI system fielded for classified settings. Human operators would then ask the system to analyze the information and prioritize targets while accounting for real-time factors like aircraft locations. Humans would remain responsible for checking and evaluating the results—but the AI would be doing the initial ranking and recommendation.

OpenAI’s ChatGPT and xAI’s Grok could theoretically be the models used for this scenario, as both companies recently reached agreements for their technologies to be used by the Pentagon in classified settings.

Two Technologies, One Targeting Pipeline

The military’s approach involves layering two distinct AI technologies. Since 2017, the Pentagon has operated Project Maven, a “big data” initiative using computer vision to analyze drone footage and identify potential targets. A 2024 Georgetown University report showed soldiers using Maven to select and vet targets, significantly speeding up the approval process.

Now, generative AI is being added as a conversational layer on top of this existing infrastructure. The official’s comments suggest chatbots could accelerate how quickly the military can find and analyze data when making targeting decisions.

But there’s a fundamental difference between these technologies. Maven’s interface forces users to directly inspect data on battlefield maps. Generative AI outputs are easier to access—but significantly harder to verify.

“Generative AI systems are much less battle-tested. And while Maven’s interface forced users to directly inspect and interpret data on the map, the outputs produced by generative AI models are easier to access but harder to verify.” — MIT Technology Review Analysis

The Accountability Gap

The use of generative AI for targeting decisions is reducing the time required in the process, according to the official. But when asked how much additional speed is possible if humans must still double-check model outputs, they declined to provide details.

This opacity is particularly concerning given recent events. Multiple news outlets have reported that Anthropic’s Claude was integrated into military AI systems and used in operations in Iran and Venezuela. While the Washington Post reported Claude and Maven were involved in targeting decisions in Iran, there’s no evidence yet explaining what role generative AI played in the school strike.

The New York Times reported that a preliminary investigation found outdated targeting data partly responsible—a reminder that even sophisticated AI systems are only as good as the information feeding them.

The Corporate Chess Game

The Pentagon’s AI partnerships have become increasingly contentious. Anthropic’s Claude was the first generative AI model approved for classified military use. But following disagreements over whether Anthropic could restrict military applications, the Defense Department designated the company a supply chain risk. President Trump demanded on social media that the government stop using Anthropic’s products within six months; the company is now fighting the designation in court.

OpenAI announced its Pentagon agreement on February 28, with stated limitations—though the practical effectiveness of those limitations remains unclear. xAI has also reached a deal for Grok to be used in classified settings.

The pattern is clear: as competition for military AI contracts intensifies, the guardrails around how these systems can be used are becoming negotiable.

What Comes Next

The Pentagon has been ramping up AI deployment across operations. Since December, millions of service members have had access to nonclassified generative AI models through GenAI.mil for tasks like contract analysis and presentation writing. But classified use for targeting represents a qualitatively different application—one with life-or-death consequences.

The fundamental question isn’t whether AI can help military decision-making. It’s whether the speed gains are worth the verification challenges, and who bears responsibility when systems optimized for efficiency produce catastrophic errors.

For now, the official would neither confirm nor deny whether this represents current operational reality. But the fact that the Pentagon is actively considering handing target prioritization to chatbots suggests we’re already further into the future than many realized.

This article was reported by the ArtificialDaily editorial team. For more information, visit MIT Technology Review.