The ask moved from compensation to code

Here's the deal: on July 28, filings surfaced showing that Jess Asato, a Labour MP in the UK, is seeking a new court order in her case against Elon Musk's xAI. What she's asking for is unusual.

Not money. Not an apology. She's asking the court to require xAI to implement "effective and permanent technical measures" — in plain terms, to make it so Grok cannot generate manipulated images of her going forward.

That matters because nearly every AI lawsuit so far has lived in the realm of after-the-fact remedy. Take the content down. Pay damages. License the training data. All of them are claims about output. What Asato's side is seeking is an order about the model's capability itself. It puts a hard question directly in front of a judge: can a court instruct a technology company to make a feature stop working?

Her own words compress the legal theory:

"Grok created deepfake pornography and sexualised content which harmed thousands of women and children. Its ability is not an accident, nor misuse, it is a design choice..."

Who's involved

Jess Asato is a Labour member of the UK Parliament. In early January 2026, content manipulating her image began circulating from Grok — starting with a fabricated bikini image and extending to a video depicting her being chloroformed and prepared for sexual assault. Public figures being targeted with this kind of material isn't new. What's different is that the production happened as a default feature of a commercial AI product anyone can use.

xAI is Musk's AI company, and Grok is integrated into X. Grok's image and video generation has been notably looser on content constraints than rival commercial models, and the company has leaned into "uncensored AI" as a positioning. That positioning is now precisely the issue in dispute, because the core of Asato's claim is that the looseness wasn't an oversight — it was intended product design.

The legal team signals how this case is meant to be read. Solicitors are AWO, a data-rights firm, with Ravi Naik as lead solicitor and Lucie Audibert and Alex Lawrence-Archer on the team. Counsel includes Marie Demetriou KC (Brick Court Chambers) plus Edward Craven KC and Rosalind Comyn (Matrix Chambers). KC is the top rank of the English bar. This is staffed as strategic litigation aimed at precedent, not a personal reputation claim.

Naik's line makes the intent explicit:

"At its heart, this case is about a single principle – that AI developers must answer for the way they design their tools."

What actually happened

Date Event
Early Jan 2026 Asato's image manipulated via Grok and circulated
Jan 12, 2026 Ofcom opens formal investigation into X under the Online Safety Act
Jan 2026 Malaysia and the Philippines temporarily block Grok (later lifted)
Jan 26, 2026 European Commission opens investigation into illegal content dissemination in the EU
Feb 17, 2026 Ireland's Data Protection Commission opens its investigation
Jun 3, 2026 Asato files claim in the High Court of England
Jul 28, 2026 Filings seeking "effective and permanent technical measures" reported

The causes of action are two: breach of data protection law and misuse of private information. The interesting choice is that this isn't a defamation case. Rather than litigating whether a manipulated image constitutes a false statement of fact, the claim frames it as her personal data — her face, her identity — being processed without consent. Under UK and EU data protection regimes, that's a far sharper instrument.

Three remedies are sought: a declaration of illegality, damages, and an order requiring preventative technical measures. The first two are conventional; the third is what the July filing focuses on. It also explains why the case continues even though the offending images have already been removed. Takedown removes this one; an order blocks the next one.

Context on scale: research from the Center for Countering Digital Hate found Grok generated roughly 3 million sexualized images in 11 days, with an estimated 23,000 depicting children. That's a volume individual remedies can't absorb, which is precisely why Asato's team pivoted toward an order about capability.

What each side stands to win or lose

What Asato's side is after is precedent beyond personal relief. Takedown is already achievable and damages here are unlikely to be large. The real objective is getting "AI developers bear legal responsibility for design choices" written into a judgment. Once that principle exists, subsequent claimants start from a much stronger position.

What xAI has to defend is more fundamental. If a court orders permanent prevention of image generation about a specific individual, that could mean maintaining per-person blocklists in filter layers or model behavior. Complete blocking is technically hard, and circumvention attempts don't stop. The company has real room to argue the order would be impossible to comply with. The claimant will answer that perfection isn't the standard — reasonable effort is. That exchange is the technical heart of the case.

Regulators gain judicial reinforcement. Ofcom has been investigating Online Safety Act compliance since January, and the EU and Ireland have opened their own tracks. Regulatory processes are slow and often end in administrative penalties; a civil judgment landing first strengthens the regulators' arguments.

But the case has a structural weakness. The claimant is a sitting MP with top-tier counsel. The overwhelming majority of people appearing in those 3 million images have nothing of the kind. Even if precedent gets set, using it requires being able to run a High Court case. The criticism that individual litigation is a poor fit for a mass-automated-generation problem is fair. This ultimately needs regulation and legislation, and the practical value of this suit may lie in supplying case law as raw material for that legislative argument.

How cases like this have gone before

Orders compelling technical measures do have precedent. Blocking orders against internet service providers are the obvious one. English courts have ordered ISPs to block access to infringing sites, and EU courts have addressed the legality of similar orders. The governing principle was that the measure need not be perfect — it needs to make infringement substantially harder. Expect Asato's team to try to carry that doctrine over to AI models.

There's also a precedent about limits. After the "right to be forgotten" ruling covering search results, enforcement proved grinding. Deletion requests exploded, operators applied their own judgment, and consistency problems persisted. It demonstrated that per-individual blocklists buckle at scale. An order against Grok would likely meet the same problem.

The most directly relevant precedent is deepfake criminal legislation. The UK has introduced law criminalizing non-consensual AI deepfake creation. But criminal law punishes the creator; it doesn't regulate the tool. Individually prosecuting tens of thousands of anonymous users isn't realistic, which is why liability is beginning to point at the tool provider. This case is the test of that shift.

One more: whether platform immunity logic applies. Platforms have historically received broad protection for content users upload. But Grok content isn't uploaded — the product generates it. That distinction is where the claimant's force comes from, and it's why "design choice" keeps recurring in her language. Whether generative AI fits inside the existing platform-immunity frame is being asked in several jurisdictions simultaneously, and this case could produce one of the early answers.

How the other players move

Ofcom and the UK government are already on a hard line. The January 12 investigation explicitly examines whether X met its Online Safety Act duties and, notably, whether it conducted the risk assessments required before significant service changes. That framing matters: asking "did you follow the process" is far easier to prove than "was the outcome bad." Fines and service restrictions have been discussed.

The EU and Ireland are approaching via the Digital Services Act and GDPR respectively. Ireland's DPC in particular functions as the lead supervisory authority for xAI's European data processing, which gives it substantial leverage.

Other frontier labs are quietly benefiting. OpenAI, Google and Anthropic have imposed much stronger constraints on generating images of real people, and those constraints drew steady criticism for being over-conservative. The Grok situation reframes that conservatism as regulatory risk management. Expect "this class of incident is structurally impossible for us" to become a sales argument in enterprise deals.

xAI's response looks two-track. One is litigating feasibility and jurisdiction. The other is product updates — Musk has said Grok 4.6 is weeks away with 4.7 to follow. If new versions ship stronger safeguards, the company can argue it acted voluntarily and no order is necessary. That's a standard litigation play.

And open-source image models remain the structural variable. Regulating commercial products doesn't reach models running locally on someone's machine. Winning this case doesn't solve the whole problem. Scale does differ, though — a default feature in an app used by hundreds of millions and a locally configured model have vastly different reach.

So what changes

If you build AI products, the practical takeaway is documented pre-launch risk assessment. Note that Ofcom's central question is whether a risk assessment preceded a significant change. When an outcome goes badly, the defense doesn't start with your benchmark scores — it starts with the fact that you assessed, and there's a record. Especially true if you're shipping image or video generation.

If you work in legal or compliance, watch the choice of cause of action. Not defamation — data protection and misuse of private information. That pairing looks likely to become the established route for personal harms caused by generative AI.

If you're an individual user, the practical state of play is this: getting existing images taken down is comparatively achievable, but preventing generation in the first place is not yet legally established. This ruling will draw that line.

If you follow AI policy, the significance is the relocation of responsibility. Deepfake response has so far meant punishing the person who made it, and that approach collapsed at scale once tools enabled mass automated generation. Whether design liability attaches to the tool provider is now on trial.

If you're just reading the news, keep the claimant's sentence: "not an accident, nor misuse, it is a design choice." This case isn't about bad users. It's about the product decisions that made that use possible.

🥄 Three Things You're Probably Wondering

— Can a court really order "turn this capability off"? There's precedent in the neighborhood. English courts have ordered ISPs to block infringing sites, under a doctrine holding that a measure needn't be perfect if it makes infringement substantially harder. But blocking access and constraining a model's generative capability aren't the same thing. xAI will argue impossibility of compliance, and that argument is the technical core of the case.

— The images were already removed. Why keep litigating? Because removal only erases what already exists. The claimant's logic is direct: leave the tool as it is and it gets made again tomorrow. The ask moved from takedown to prevention, and that shift is the whole case.

— Don't other AI systems have this problem too? Degree differs. Most major commercial models impose strong constraints on generating sexual imagery of real people — constraints they've been criticized for as too conservative. Grok became the target because loose constraints shipped as a default in an app with hundreds of millions of users. That said, regulation can't easily reach open models running locally, and that problem survives whatever happens here.

Sources

The case is ongoing; facts and procedure may change as the court rules.