Every AI company that builds image generators — tools that create pictures from text descriptions — ships them with safety filters. These are automated guardrails that block the system from producing harmful content: violence, hate speech, sexual imagery of real people. OpenAI has them. Google has them. Midjourney has them. It's the bare minimum. The "please don't get us sued" baseline.

Elon Musk's xAI apparently missed that memo.

On March 24, Baltimore filed a lawsuit against X Corp., x.AI Corp., x.AI LLC, and SpaceX in the Circuit Court for Baltimore City. The allegation: Grok — xAI's generative AI system (software that creates new content from text prompts) — produced an estimated 3 million sexualized deepfakes in just 11 days around the turn of 2026. Deepfakes are AI-generated fake images or videos designed to look indistinguishable from real photos. More than 23,000 of those images depicted children.

Three million. Eleven days. Twenty-three thousand children. The Center for Countering Digital Hate produced these numbers, and Baltimore's complaint cites them directly.

The specific mechanism? Grok's so-called "spicy mode" — a feature that, according to the lawsuit, allowed users to ask Grok to undress or "nudify" photos of celebrities, private citizens, and minors. Despite xAI's public claims that such content was prohibited, Grok routinely generated it with minimal prompting.

Baltimore — population roughly 568,000, now the largest US city to take legal action against xAI — isn't playing nice. The city wants the highest possible financial penalties, a court-ordered injunction (a legal order forcing xAI to redesign Grok's content filters), and changes to how xAI markets the platform. SpaceX got named as a defendant due to shared corporate infrastructure.

The legal strategy is the real story

Baltimore isn't waiting for Congress to draft new AI-specific legislation. The city is using its existing Consumer Protection Ordinance — a local law that protects residents from unfair and deceptive business practices. The argument: Grok's design itself is the problem. Not any single image. The entire product is defective.

Here's the analogy that makes this click: if a blender company sold a product that exploded in 23,000 kitchens in 11 days, nobody would demand new "blender legislation" before suing. Existing product liability law already covers defective products. Baltimore is applying the same logic to AI. If this case succeeds, any city with a similar consumer protection ordinance gets a ready-made legal template. No waiting for Congress. No lobbying. Just a lawyer and a filing fee.

And this goes beyond civil court. Federal laws like the PROTECT Act already criminalize computer-generated child exploitation material — even AI-generated content. If the 23,000 figure holds up, xAI could face criminal referrals on top of the civil lawsuit. That's not a fine. That's a whole different category of problem.

Baltimore isn't alone, either. On March 16, three teenagers in Tennessee filed a proposed class-action lawsuit against xAI after Grok generated sexualized images depicting them — images created from school yearbook photos and social media pictures run through "spicy mode" via a third-party app. And xAI faces regulatory probes across the Americas, Europe, Asia, and Australia.

Why every AI company's legal team just had a bad Monday

Every company building image generation tools faces the same question: how do you ship a competitive product while preventing the exact misuse Baltimore alleges? The answer is embarrassingly obvious: content moderation. The thing every competitor already ships. The boring, expensive, unglamorous engineering work that doesn't make for a good product launch tweet. OpenAI did it. Google did it. Midjourney did it. xAI either underinvested or made a deliberate choice to skip it.

Musk's "free speech absolutist" brand just collided with a hard wall: generating non-consensual sexualized images is not free speech. Generating such images of children is a federal crime. The fact that this requires litigation to establish says more about xAI's priorities than about the legal system.

The verdict

This case will likely end in a massive settlement or court-mandated redesign of Grok's safety systems. Either way, Baltimore is setting a precedent. AI companies that treat safety engineering as optional will learn the lesson the expensive way — from judges, not engineers.

The "move fast and break things" era of AI image generation just discovered that some things, once broken, come with a court date.