OpenAI Moves Into US Military Systems as xAI Is Sued Over Grok and Images of Teens

In its March 17, 2026 issue, MIT Technology Review paired two stories that point to the same shift: generative AI is entering sensitive government environments while failing, more visibly, where basic safeguards are needed. OpenAI is preparing to work in the Pentagon's classified systems, and xAI has been sued over Grok, which, according to the plaintiffs, helped create child sexual abuse material (CSAM) from real photos of teenagers.

Where OpenAI will show up

On February 28, 2026, OpenAI announced an agreement with the Pentagon to deploy its models in classified environments. The company separately stresses three "red lines": no mass domestic surveillance, no use of its models to direct autonomous weapons, and no high-stakes automated decisions without a human involved.

On March 2, OpenAI added language to the contract that explicitly bans domestic surveillance of US citizens and bars use of the service by agencies such as the NSA without a new agreement.

But the main question is not the wording; it is how the technology will be used in practice. According to MIT Technology Review, the integration is not yet complete: OpenAI's models still have to be built into existing military systems and processes. Against that backdrop, it is striking how quickly the company turned toward the defense market after the Pentagon's clash with Anthropic. Where military use of AI was once discussed in terms of future risks, it is now a concrete deployment, in the midst of the conflict with Iran.

Three use cases

MIT Technology Review describes several areas where OpenAI could appear in US military infrastructure in the near term.

• Target prioritization. A military analyst could upload a list of potential targets, logistics data, video and images, and the model would return recommendations on what to check and what to strike first. Formally, a human keeps the final decision, but this is where the question arises of how far "manual control" really restrains the speed of machine suggestions.
• Air defense and counter-drone work. Through a partnership with Anduril, OpenAI's models could serve as a conversational layer on top of systems that already analyze sensor data and battlefield video. The AI would not necessarily aim weapons itself, but it could offer soldiers an interpretation of threats and response options in natural language.
• Military bureaucracy. OpenAI's models are already slated for the GenAI.mil platform, where they are meant to draft contracts and policy documents and provide administrative support for missions. This is a less dramatic scenario, but it shows how AI is becoming part of the Pentagon's everyday work.
• Rapid scaling. If the models prove useful, they could be built fairly quickly into a broader stack of military interfaces and services. That raises not only potential efficiency but also the scale of errors.

The lawsuit against Grok

At the other end of the same story is not a battlefield but a consumer product without sufficient restrictions. On March 16, 2026, three teenagers from Tennessee filed a lawsuit against xAI in federal court in California. According to the plaintiffs, Grok's tools were used to create sexualized images of minors from ordinary photos in which they were clothed.

The case materials mention photos from graduation events, a school yearbook and social media; according to the complaint, these files were then distributed through Discord and Telegram and traded for other illegal content. The plaintiffs want the case certified as a class action and say they would represent, in the complaint's words, thousands of potential victims.

The lawsuit also alleges that xAI knowingly released a system capable of generating explicit content even though the risk of CSAM was foreseeable. Grok had already been caught up in scandals over sexualized deepfakes, and after the January wave of criticism some features were made available only to paying subscribers, which, according to the plaintiffs, did not solve the problem but merely monetized it.

"These children may now live their whole lives with the fact that such

images exist on the internet," the plaintiffs' attorney said, describing the consequences of the case.

What this means

These two stories show that the AI industry is facing a double crisis of trust. On one side, models are moving deeper into military and government decisions, where the cost of error is measured in lives and political consequences. On the other, even basic consumer products still cannot reliably block obviously dangerous scenarios.

The signal for the market is simple: the debate around AI is no longer about demos and wow effects, but about responsibility, control and the real limits of what is acceptable.