A former engineer at xAI, Elon Musk's AI company, has sued the firm in California state court claiming he was fired after raising repeated concerns about safety problems in Grok, the company's AI chatbot. The lawsuit opens a new legal chapter for xAI, which is already facing multiple other court battles tied to its flagship product.

Devin Kim filed the complaint on June 10, 2026, according to reports by Law360 and other outlets. Kim says company leaders were warned multiple times about risks in Grok before he was let go, which he characterizes as illegal retaliation. He chose to file in California state court, which has strong whistleblower protection laws that make it harder for employers to defend themselves—a choice that likely reflects his legal strategy.

What Kim's Complaint Alleges

Kim says he flagged two categories of concern. First, he identified discrimination risks—bias built into the AI model through skewed training data or incomplete testing. Second, he raised broader public safety issues. According to reports from Yahoo Finance and the Daily Journal, his warnings were not only ignored but ultimately cost him his job.

The bias problem Kim flagged is a known technical challenge. Language models absorb biases from the internet data they learn from—if that data reflects human prejudices, the model will too. Companies working on safety typically try to address this at multiple stages: during initial training, during fine-tuning (where engineers shape the model's behavior), and through "red-teaming" (internal testing that tries to break the model). An engineer raising these issues formally is doing what safety-conscious companies say they want. Whether xAI treated that as helpful input or as a problem to be silenced is now for the courts to decide.

xAI has not issued a public response to Kim's specific allegations as of now.

A Second Front: Baltimore's Deepfake Case

Kim's suit does not stand alone. In March 2026, the City of Baltimore filed its own lawsuit against both X and xAI. The city alleges that Grok was used to create non-consensual sexual deepfakes—fake images of real people in sexual situations without their consent—and that the companies failed to build in safeguards to prevent this abuse. Details were published in litigation documents by DiCello Levitt.

This is a specific, documented form of harm. An AI generates a fake image; the victim is identifiable but unaware; the image is shared without consent. From a technical angle, the suit raises a design question: how well does Grok's safety system work at runtime—that is, when users are actually using it? Can people find ways around the safety filters, and if so, does the company detect and fix those workarounds?

At its core, Baltimore's argument is that there was a gap between what xAI said publicly about safety and what Grok actually did in practice—and that the company either knew or should have known about this gap.

If Kim's account is accurate, his internal warnings could strengthen Baltimore's legal position. Courts and juries tend to take notice when a company was warned internally about a problem, did nothing, and then that exact problem caused real-world harm.

A Third Case: The Trade Secrets Lawsuit

This is where the legal picture becomes more complex. xAI itself filed suit against a different former engineer. In August 2025, Reuters reported that xAI accused Xuechen Li of stealing company files and trade secrets related to Grok. According to xAI's complaint, Li admitted in a meeting on August 14 that he had taken files and tried to cover it up. The suit alleged the material was headed to OpenAI, one of xAI's biggest competitors.

The Li case and the Kim case point in opposite directions legally. In the Li case, xAI is suing; in the Kim case, xAI is being sued. But together they paint a picture of a company managing two difficult internal challenges at the same time: protecting its intellectual property and managing safety governance questions. Both have turned into public court battles within a year.

A Pattern in Technology Company Growth

I've spent three decades watching technology companies scale, and moments like this follow a recognizable pattern. The setup is usually: rapid hiring, a product under intense competitive pressure, and internal processes that haven't caught up to the company's size. We saw versions of this in the late 2000s and early 2010s when cloud companies and social platforms were growing fast. Companies would fight IP theft lawsuits with one hand while employment disputes and regulatory complaints came in through the back door. xAI appears to be living through a compressed, more intense version of that same dynamic—compressed because the stakes in AI are unusually high, and the public and regulatory scrutiny is sharper.

The companies that handled those earlier transitions better did something specific: they treated safety and IP protection as real infrastructure, not just as a box to check for compliance. And they listened when an engineer inside raised a documented concern—because that signal usually mattered.

What This Cluster of Cases Means

Three active or recent lawsuits involving Grok—one over firing a safety whistleblower, one over deepfake harm, and one over stolen trade secrets—do not by themselves prove wrongdoing. Courts will decide each case based on its own facts and evidence. But for a company as young as xAI, facing legal exposure across employment law, product safety questions, and intellectual property protection is unusual. It arrives at a moment when regulators worldwide are getting stricter about AI safety obligations.

For legal and governance professionals tracking AI risk, the Kim case raises a crucial question that will ripple through the whole industry: will California courts treat internal warnings about AI safety as protected speech under existing whistleblower laws, or will the lack of AI-specific language in those laws create enough gray area to favor the employer? How that question gets answered will affect not just xAI but many AI companies.

The Baltimore case is part of a broader wave of lawsuits filed by cities and states targeting AI-enabled image abuse. If those suits succeed, they could create a template for holding AI developers financially responsible for how their systems are misused at the point of use—regardless of what users do with the technology.

One important caveat: none of these cases has been decided yet. Every claim remains an allegation unless a court rules otherwise. xAI has not publicly responded to Kim's complaint, and the company's legal position on the Baltimore suit has not been widely reported.

The Grok Product at the Center

Grok has been updated several times since xAI was founded. The company marketed it as having access to real-time information through its connection to X's massive data feed, and positioned it as less restricted than models from OpenAI, Anthropic, and Google. That positioning created a built-in tension: you cannot claim to be "less restricted" while also meeting strict safety standards. That tension—and how xAI manages it—is what the Kim and Baltimore cases are now asking in open court.