Pentagon: Generative AI at the Heart of Military Targeting Decisions

Generative AI: A New Tool for Military Targeting

The U.S. military is considering integrating generative AI systems into its targeting operations, according to a Defense Department official. These systems could be used to classify target lists and recommend which ones to prioritize for attack, while ensuring that final decisions are made by humans. This revelation comes as the Pentagon faces criticism following a strike on an Iranian school, which is currently under investigation.

The proposed process would involve inputting a list of potential targets into a generative AI system deployed in classified environments. Humans could then ask the AI to analyze this information and prioritize targets based on various factors, such as the location of aircraft. The AI's results and recommendations would then be verified and assessed by humans. Models like ChatGPT from OpenAI and Grok from xAI could be utilized in this type of scenario in the future, as these companies have recently entered agreements for their models to be used by the Pentagon in classified settings.

Strategic Partnerships with OpenAI and xAI

Although the official did not confirm whether this method is currently in use, he described it as a potential example of operation. Other media outlets have reported that Claude from Anthropic has been integrated into existing military AI systems and used in operations in Iran and Venezuela. The official's comments shed light on the specific role that chatbots could play, particularly in accelerating target research. They also highlight how the military is deploying two different AI technologies, each with its own limitations.

Since at least 2017, the U.S. military has been working on a "big data" initiative called Maven. It employs older types of AI, particularly computer vision, to analyze the vast amounts of data and imagery collected by the Pentagon. Maven could, for example, take thousands of hours of drone aerial footage and algorithmically identify targets. A 2024 report from Georgetown University showed that soldiers were using the system to select and verify targets, which expedited the approval process for these targets. Soldiers interacted with Maven through an interface featuring a battlefield map and a dashboard, which could highlight potential targets in one color and friendly forces in another.

The Maven Initiative and the Evolution of Military AI

The official's comments suggest that generative AI is now being added as a layer of conversational chatbot—one that the military could use to find and analyze data more quickly when making decisions about which targets to prioritize.

Generative AI systems, like those underlying ChatGPT, Claude, and Grok, represent a fundamentally different technology from the AI that has primarily powered Maven. Based on large language models, they are much less battle-tested. While Maven's interface required users to directly inspect and interpret data on the map, the results produced by generative AI models are more accessible but harder to verify.

Using generative AI for such decisions reduces the time needed for the targeting process, the official added, though he did not provide details when asked how much additional speed could be gained if humans had to spend time verifying the results of a model.

Consequences and Controversies

The use of military AI systems is under increased public scrutiny following the recent strike on a girls' school in Iran, where over 100 children lost their lives. Several media outlets reported that the strike was carried out by an American missile, although the Pentagon stated that an investigation is still ongoing. And while the Washington Post reported that Claude and Maven were involved in targeting decisions in Iran, there is still no evidence explaining what role generative AI systems played, if any. The New York Times reported Wednesday that a preliminary investigation found that outdated targeting data was partly responsible for the strike.

The Pentagon has ramped up its use of AI in its operations in recent months. It has begun offering unclassified use of generative AI models for tasks like contract analysis or presentation drafting to millions of service members since December, through an initiative called GenAI.mil. However, only a few generative AI models have been approved by the Pentagon for classified use.

The first was Claude from Anthropic, which, in addition to its use in Iran, was reportedly used in operations to capture Venezuelan leader Nicolas Maduro in January. However, following recent disagreements between the Pentagon and Anthropic regarding the company's ability to restrict the use of its AI by the military, the Defense Department designated the company as a supply chain risk, and President Trump demanded on social media that the government cease using its AI products within six months. Anthropic is contesting this designation in court.

OpenAI announced an agreement on February 28 for the military to use its technologies in classified environments. Elon Musk's company, xAI, also reached an agreement for the Pentagon to use its Grok model in such settings. OpenAI stated that its agreement with the Pentagon comes with limitations, although the practical effectiveness of these limitations remains unclear.