“AI” models left trying to fix the mess

By Alicia Lucas

Part 1: “AI” models left trying to fix the mess – better to try and stop it during the early stages

Litigation, contesting whether something has been done in accordance with the law, in the USA is expensive. The powerful and well-off appear most likely to have their lawsuits, their complaints about a perceived wrong, upheld while everyone else ponders whether they can even afford the cost of a day, let alone a full trial. The widespread alleged harms and damages, including facilitating murder, that people have experienced with the arrival of “AI”-labelled models (machine learning models) and associated infrastructure such as data centres, has spurred individuals, communities and businesses to file lawsuits with the help of, what seem to be generally sympathetic, lawyers – despite the potential costs.

Lawsuits are a way for these people who feel invisible to have a voice. Generally, no notice beforehand, or chance to comment, was given to people potentially effected by what these companies were doing. Possible negative effects on people living near where companies wanted their infrastructure or people using the models or their families and friends, largely, didn’t seem to be a priority. People are now trying to fix this mess.

Its been around four years since the implementation of “AI”-labelled models and a broad range of suits hav been filed. Models include OpenAI’s ChatGPT (CEO Sam Altman), Anthropic’s Claude (CEO Dario Amodei), Google’s Gemini (CEO Sundar Pichai) and Space X (X.AI)’s Grok (CEO Elon Musk) and similarly built companion models like Character.ai. Harms and damages including, allegedly, false advertising, disadvantaging people and companies by taking their knowledge without permission, and facilitating death. Only some cases are mentioned in this article. Part 1 doesn’t include cases in regards to mental health delusions and similar, suicides or murder.

Most cases were chosen because previous lawsuits showed some glimmers of light. That is, there were partial successes and companies addressed wrongs to some degree. While the focus is on lawsuits in the USA, some description of relevant happenings outside of the USA are mentioned, including from Australia. The case immediately below about a lawsuit against a town, perhaps, provides an example of how lawsuits may be used by the wealthy to gain what they want while the second case, initiated by a community group in regards to illegal powering of data centres, is one the USA Federal Government is attempting to stop from proceeding.

Data Centre lawsuit against a town

Developer, Related Digital, and 3 property owners from Saline Township, Michigan, USA, filed a lawsuit on the 12th September 2025 against the Saline Township Board when the Board voted 4-1 against rezoning 575 acres (233 ha) of land from “farmland” to “industrial”. The developer wanted to build a data centre – it needed to be industrial land. The “packed township hall” was described as “overwhelmingly opposed to the data centre.” Saline Town is a small community of about 2 200 people. Related Digital was formed earlier in the year by real estate firm Related Companies. Stephen Ross, described as a billionaire, is the founder and non-executive chair of this company. The Town Board settled with Related Digital and the property owners near the end of October 2025 and the rezoning went ahead. On the 1st November 2025 it was reported that Oracle and OpenAI had announced they would build a 1 gigawatt-plus data centre on the land in conjunction with Related Digital. (The Saline Post) (Planet Detroit) (AP News).

Data Centre lawsuits and the USA Federal Government’s Intervention

Notice was given in June 2025 and a lawsuit filed on the 14th April 2026 alleging X.AI Corp and MZX Tech LLC use without permits gas power plants for their “Colossus” data centres that power the company’s data centres in the greater Memphis area. Permits are required under the Clean Air Act 1977. MZX Tech LLC is a subsidiary of X.AI, which is now a subsidiary of SpaceX. The Southern Environmental Law Center [sic] filed the lawsuit on behalf of the NAACP. The NAACP seeks to end race based discrimination and acts “for  the civil rights due to Black America”. These gas turbines are located where Black people comprise a large proportion of the community. There are 3 data centres, Colossus 1 to Colossus 3. It is not clear how 3 is being powered. The Grok model was said to be the reason these data centres needed to be built but both Google in June 2026 (Colossus 1) and Anthropic in May 2026 (Colossus 1 and 2) have leased space to run models or use  for cloud services. It has been reported X.AI are using Colossus 2. (NAACP Press Statement) (United States District Court, Oxford Division, NAACP v X.AI Corp, Case 3:26-cv-00074, Doc: 1, Filed 14/4/26).

On the 8th June 2026 a class action lawsuit by residents was filed against X.AI, MZX Tech LLC and Space Exploration Technologies Corp (aka SpaceX) to try and address the allegedly increasing noise from the gas power plant used to supply energy for the company’s data centres, large industrial facilities needing lots of energy to operate. Gas-fired turbines have risen in number from 3 to 57 in less than a year. The company has allegedly done nothing to try and address the noise. Turbines are described as ‘temporary”but continue to be used (although the company has now received permission to build 41 permanent turbines). They run largely non-stop and residents say the noise is like that of a “jet engine” and can include a “combination of high-pitched squealing, continuous engine roaring, low-frequency rumbling, and tonal humming or whining … It can also be felt as vibration.” (United States District Court, Oxford Division, Haley v X.AI, Case 3:26-cv-00148, Filed 8/6/2026).

Residents were given no opportunity to understand and voice concerns prior to learning from news reports that MZX Tech had purchased the property that would be used only for a gas power plant.

On the 15th June 2026, the USA’s federal Department of Justice sought to object to the lawsuit and request its dismissal as the data centres used for the “AI”-labelled Grok model support the Department of War’s military operations. SpaceX is one of the companies now able to undertake classified work for the military. None of the land was defence land prior to use By X.AI. (United States District Court, Oxford Division, NAACP v X.AI, Case 3:26-cv-00074, Doc: 59, Filed:15/06/26)

The land to hold data centres was leased or bought by X.AI through a subsidiary known as CTC Property LLC. This meant the community wasn’t necessarily aware that an “AI”-labelled company was the leaseholder/purchaser and may want to build large data centres on this land. The land for Colossus 1 was leased in March 2024 while the land for Colossus 2 was bought in March 2025. The owners of the Colossus 1 land sold it to a subsidiary of SpaceX in May 2026.

Subscription costs

A potential class action submitted to the United States District Court, San Francisco Division, on the 15th June 2026, concerns Anthropic’s allegedly “misleading and deceptive marketing of its Claude Max subscription plans.” Two of the company’s three individual subscription plans were labelled in a way that lead purchasers to believe they were gaining significant more power than could be obtained using the other plan. Subscribers to either of the two more powerful plans have allegedly found they are getting far less then they expected. The two more powerful plans are not cheap. One plan costs US $100/month and the other is US $200/month. (United States District Court, San Francisco Division, Karl Kahn v. Anthropic, PBC dba Anthropic, Inc, Class Action, Case No.: 3:26-cv-5763).

Of course, this may be thrown out of court but, at the least, it may indicate Anthropic’s customer information needs to be improved so buyers can be aware, not confused.

Illegal use of copyrighted books and scholarly articles (not newspapers)

Writers have been negatively impacted by the development of “AI”-labelled models. Their words are considered useful to infuse the models with such things as quality writing, context and high-level scientific understanding. Copyright holders – authors and publishers – have tried to seek redress for companies involved in “AI”-labelled model building taking copyrighted works without permission. Part of a lawsuit against Anthropic indicated some redress was possible. A judge ruled, in June 2025, the use of works was legal under USA copyright law, however, the judge found this was not the case when they were pirated copies. Anthropic agreed to settle the lawsuit in August 2025, although not admitting fault, and pay authors about USD $1.5 billion. (United States District Court, Northern Division of California, Bartz v. Anthropic PBC, No. 3:24-cv-05417, Document 231).

New class action lawsuits are being pursued. Five publishers of literature, educational books and scholarly articles and a best-selling author filed a class action lawsuit on 5th May 2026 against Meta and CEO Mark Zuckerberg for inappropriate use of millions of textual works to develop the “AI”-labelled Llama model (see Hachette below). This lawsuit is also being taken on behalf of a proposed class of analogous copyright owners. Similarly, Anthropic as well as CEO Daniel Amodei and co-founder Benjamin Mann face a class action lawsuit filed on the 17th June 2026 for deliberately using illegal versions of the works of many authors. (United States District Court, Northern Division of California, Shakespeare v. Anthropic PBC dba Anthropic, Inc. Case No. 3:26-cv-5931).

Law-and made up law

It is hard to understand how lawyers trained to understand and up hold the law could rely on computer models such as ChatGPT or Claude. A well known saying in regards to statistical and mathematical modelling that may also be useful for lawyers is “All models are wrong, but some are useful” by George E.P. Box. This saying doesn’t mean a model can’t give a correct answer more that it is just a model and not an exact replica of the real thing. All of the large models like ChatGPT, Grok and Claude can provide wrong answers and some are anthropomorphically mislabelled as “hallucinations”. Experts in machine learning have said “hallucinations” are impossible to eliminate entirely from answers given by a model. In July, 2025, a lower court’s ruling that relied on false citations generated by an “AI”-labelled model was thrown out by the Georgia Court of Appeal, USA. (FindLaw: Shahid v. Esaam 2025) In May, 2025, a lawyer representing Anthropic in a lawsuit regarding music copyright produced a made-up citation (the article, itself, was correct) that came from the company’s Claude model. (Reuters) Rather than take this limitation of “AI”-labelled models to heart it appears lawyers have just kept generating on. In Florida, Fourth District Court of Appeal Chief Judge Mark Klingensmith said, as quoted in the Florida versus OpenAI, CEO lawsuit, in 2026 “[in] fact, the number of filings our court has seen containing AI misuse has increased over time.” (Circuit Court of the Tenth Judicial Circuit, Highlands County, Florida, Filing #249302659).

Lawyers and self-represented litigants must now certify legal authorities cited in filings are accurate the Florida Supreme Court has ruled. Judges can sanction those who submit “filings containing fabricated or inaccurately cited authorities, including citations generated by AI tools” The Florida Bar reports. (Circuit Court of the Tenth Judicial Circuit, Highlands County, Florida, Filing #249302659).

Sexual exploitation of women and children

Nonconsensual deepfake images can still be made in XAI’s “AI”-labelled model Grok and posted in X according to Wired magazine as reported in Social Media Today in June 2026. That’s about five months after XAI was forced to address this functionality after rapid and widespread outcry about how harmful it was.

X.AI gave “AI”-labelled model Grok users the ability to alter a publicly available image they had uploaded by simply describing the desired changes in December 2025. The altered image could then be posted to X for any user to see. Images of clothed, real women were digitally undressed and put in skimpy bikinis and real children were shown in explicit sexual poses amongst many other sexualised images posted on X. These changed images are known as “deepfakes” as they are difficult to distinguish from real images. Animated images could also be made. It has been reported that if a women publicly complained about an image on X, a barrage of new altered images would appear. The functionality had previously been released as a standalone app in August 2025 and non-profit groups such as Consumer Federation America had immediately labelled the product “dangerous and violative” and requested these problems with the app be addressed. (Democrats, Energy and Commerce Committee 2026).

In early January, 2026, increasing complaints about the sexualised images came from affected women and others including public officials from the UK, France and India. X.AI responded by restricting access to the image editor to premium customers only. More serious action was taken against the company. The first lawsuit was filed, Indonesia (see BBC 12/1/2026), Malaysia and the Philippines blocked access to Grok, the UK communications regulator commenced an investigation into the Grok model under the Online Safety Act and the California Attorney General Rob Bonta (Democrat) issued a cease-and-desist order. The latter was sent on the same day the BBC (16/1/2026) reported on an announcement on X that said technical changes had been implemented to prevent the “Grok account from allowing the editing of real people in revealing clothing”. The company also started to remove deepfakes around this time. (EuropeSays: Every Grok Deepfake Lawsuit and Ban in 2026).

Not long after the technical changes were made, Malaysia and the Philippines reinstated access to Grok. Indonesia did as well saying it was not an “unconditional lifting of restrictions” and monitoring of the Grok model would continue. Indonesia had also received written confirmation from X.AI that they would conform to Indonesia’s content moderation and child safety laws amongst others and received further reassurance that X.AI would “not violate privacy or public dignity.” (Indonesian National Police: Ministry normalizes Grok AI Access).

XAI continued, however, to face more investigations and legal challenges. These included: three Tennessee children filing a class-action lawsuit: the City of Baltimore, concerned the Grok model violated consumer protection law, filed a lawsuit; Jess Asato, British MP, had a number of sexualised images made of her that she did not consent to and is taking the company to the UK High Court; the European Union has banned “AI systems” or “nudifier apps” that generate child sexual abuse material or sexualised images, videos and audio of people without their consent; (News European Parliament) and; the Canadian Privacy Commissioner’s investigation into the Grok model found the image editor was implemented without evaluation of potential privacy concerns prior to release, appropriate protections were lacking after it was released and the country’s existing privacy law was violated. (News Release-Privacy Commissioner of Canada) (EuropeSays: Every Grok Deepfake Lawsuit and Ban in 2026).

Australia’s eSafety Commissioner has acted on these types of “nudify” services since late 2025. In regards, to X.AI and Grok, the company has been expected by the Commissioner to find and remove any sexualised material harmful to children and, by 6/1/2026, the eSafety Commissioner had written to X.AI to find out about the protections for children the company had put in place. Codes, drafted by industry, to protect children from such harms as sexualised deepfakes came into effect in March 2026. The eSafety Commissioner can fine social media companies up to $49.5 million dollars if they don’t comply with the codes. The codes also protect children from harms associated with chatbot models. The government may still ban “nudify” services. In regards to adults, the eSafety Commissioner can be contacted to have an image removed from a site and can investigate alleged image-based abuse of both children and adults. The eSafety Commissioner, amongst other things, does not search for sexualised deepfakes themselves and cannot charge a person with a criminal offence. (eSafety Commissioner).

The Australian Government announced on the 28th June 2026 it would seek to, not only significantly increase the fines a social media company could pay, but also give the eSafety Commissioner stronger enforcement and compliance powers.

OpenAI’s ChatGPT model has been also recently shown to generate sexualised images (BBC 18/1/2026). Researchers, who test models for weaknesses so they can be fixed, were able to tweak a readily available “prompt” and the model produced, not only, sexualised images but also very violent ones. While the sexualised images weren’t deepfakes, earlier research, explained the researchers, had shown ChatGPT could be manipulated to produce deepfakes by including images of people’s heads on an otherwise model generated image. A further concern is the actual violent and/or sexualised content of the images was produced by the model without specific instruction. The company added further “guardrails” to reduce the possibility of producing these type of images but, in even this latest version of ChatGPT (GPT 5.4), cannot guarantee complete elimination.

Continued tomorrow…

References (except court proceedings which are in text)

The Saline Post – Related digital, landowners sue Saline Township for rezoning denial.

Planet Detroit – Data center developer takes Saline Township to court over rezoning decision 18/9/2025

AP News – Tech giants announce $7B data center, Michigan’s first hyperscale campus 1/11/2025

NAACP Press Statement – Elon Musk’s xAI threatened with lawsuit over air pollution from Memphis data center, filed on behalf of NAACP 17/6/2025

Hachette, one of the plantiffs, contains a link to the complaint – Hachette book Group Publishers and authors file class action lawsuit against Meta and Zuckerberg for Willful copyright infringement to develop llama AI models.

FindLaw: Shahid v. Esaam 2025 – from FindLaw CaseLaw, Caselaw Resources, US Federal and state case law

Reuters – Anthropic’s lawyers take blame for Ai ‘hallucination’ in music publishers’ lawsuit. 16/5/2025

Democrats, Energy and Commerce Committee 2026 – E&C democrats investigate elon musk’s grok for spreading rampant non-consensual, sexualized content

EuropeSays: Every Grok Deepfake Lawsuit and Ban in 2026: UK MP joins growing legal fight against X.AI  6/6/2026

BBC – Malaysia and Indonesia block Musk’s Grok over explicit deepfakes 12/1/2026

Indonesian National Police: Ministry normalizes Grok AI Access following X corp’s compliance commitment 1/2/2026

News European Parliament – AI Act: EP approves simplification measures and “nudifier” app ban 2026

News Release-Privacy Commissioner of Canada – Privacy Commissioner of Canada investigation into the Grok chatbot and sexualised deepfakes finds companies violated privacy law 11/6/2026

BBC – X to stop Grok AI from undressing images of real people after backlash 16/1/2026

eSafety Commissioner website

Yahoo News – ChatGPT can be made to generate sexualised and violent images, researchers find 18/6/2026