Stargate to nowhere, Gen Z has little hope for AI, Musk wants the right to discriminate, AI’s mirage reasoning.

Welcome to a new edition of misaligned bits, the (roughly weekly) newsletter from Misaligned where we sum up recent news and research, sometimes with a lighter touch.

As usual, we will mark all non-medium links with “➚” (external link) and all possibly paywalled links with “🔒”.

Stargate to nowhere

OpenAI has announced it is “pausing” its Stargate Data Centre project in the UK, citing the high cost and the “regulatory environment”, by which OpenAI appears to refer to copyright legislation. The company said the project “will move forward when the right conditions such as regulation and the cost of energy enable long-term infrastructure investment.” (➚Bloomberg) While energy cost has been rising in most countries due to the war in Iran, it feels strange OpenAI did not know about the “regulatory environment” back in September when they announced the project, as nothing has changed since then? The decision is however a setback for the ambitious growth plans of the UK government.

This week, OpenAI also cancelled its plans for a Stargate data centre in Norway. After Texas and UK, this is the third data centre project Open has been putting on ice. Microsoft is now reportedly planning to take over the project (CNBC).

Meanwhile, legislators in Maine have passed the first statewide data centre ban. The governor of Main, Janet Mills, has indicated she supports the legislation. The state’s legislation will prohibit construction of data centres using at least 20 megawatts of power until late 2027 so that environmental and economic impacts could be properly evaluated.

Running dry in Memphis and more lawsuits

Elon Musk’s xAI (now part of SpaceX) has announced that its wastewater recycling plant in southwest Memphis is on hold “for the moment”. In a response the mayor of Memphis said he “spoke with a number of company reps pushing to ensure xAI delivers on the commitments made to this community”. The community group Protect Our Aquifer has called for the city to hold a public hearing. (➚WREG)

In the meantime, the civil rights organisation NAACP has filed a lawsuit against the same “Colossus” data centre in federal court in Mississippi (the data centre is located only 2 km from the state border). The complaint alleges that xAI is violating the Clean Air Act due to emissions from its makeshift power plant by operating them without sufficient permits: “Had Defendants obtained a permit, they would have been held to this standard. By simply not applying for — or receiving — a permit, xAI and MZX Tech have avoided subjecting these twenty-seven unpermitted turbines to a BACT determination.” (Case 3:26-cv-00074-MPM-JMV)

Plaintiffs are asking the court for seek declaratory and injunctive relief to enjoin xAI from operating the Colossus Gas Plant, including operating any of the twenty-seven gas-fired combustion turbines onsite.

Meanwhile, xAI has filed a lawsuit challenging Colorado’s AI bill which imposes protections against “algorithmic discrimination” in AI systems. (➚🔒Financial Times) The lawsuit claims that the legislation is “unconstitutionally vague” and violates developers’ first amendment rights.

Three weeks ago three minors from California filed a lawsuit against Elon Musk’s xAI (now part of SpaceX) over Grok’s image generation feature: “Plaintiffs are three of the minor victims of xAI’s knowing production, possession, and distribution of AI-generated child sexual abuse material (CSAM) depicting Plaintiffs.” Plaintiffs are asking the court to certify a class. (➚Case 5:26-cv-02246-SVK)

Gen Z has little hope for AI

A new study by Gallup has found that young adults have grown less hopeful and more angry about artificial intelligence. (➚New York Times) While more than half of Gen Z-ers in the United States use generative AI regularly, their feelings about the technology are souring. The percentage of respondents (ages 14 to 29) who said they felt hopeful about A.I. declined sharply since last year, down to 18% from 27%. The fall somewhat reflects the wider backlash against the AI industry. AI has been polling badly recently, which one poll showing that AI is less popular in the US than ICE (➚The Verge)

Lawsuits against OpenAI

A stalking victim has filed a lawsuit against OpenAI, alleging the company’s technology enabled the acceleration of her harassment. She claims OpenAI ignored three separate warnings that a user posed a threat to others, including an internal flag classifying his account activity as involving mass-casualty weapons. (➚TechCrunch)

Meanwhile, District Judge Seeborg has ordered that OpenAI, its CEO Sam Altman, certain employees, and investors must defend a federal court lawsuit alleging a man’s ongoing interactions with the artificial intelligence platform led him to kill his mother and himself. Seeborg denied OpenAI’s motion for dismissal. (➚Case 3:25-cv-11037-RS)

Regulatory Bits

While the UK still lacks any overarching regulation of AI, and the Trump administration is fighting actively against regulatory oversight, Nigeria has moved to introduce AI regulation modelled after the EU AI Act: The bill would introduce a risk-based framework for AI systems, requiring that they be designed and deployed in a manner that is fair, transparent, secure, non-discriminatory and subject to human oversight. (➚IAPP)

Science Bits

A team from Stanford University has found that AI models can fake visual understanding of images that don’t exist. They found that when AI models were asked about a medical image that has not been uploaded, they confidently spun detailed imaginary descriptions: “Frontier models readily generate detailed image descriptions and elaborate reasoning traces, including pathology-biased clinical findings, for images never provided.” The researchers term this phenomenon “mirage reasoning”.

Even more worryingly “without any image input, models also attain strikingly high scores across general and medical multimodal benchmarks, bringing into question their utility and design. In the most extreme case, our model achieved the top rank on a standard chest X-ray question-answering benchmark without access to any images”.

The researchers come to the conclusion that “these findings expose fundamental vulnerabilities in how visual-language models reason and are evaluated”.

Mohammad Asadi, Jack W. O’Sullivan, Fang Cao, Tahoura Nedaee, Kamyar Rajabalifardi, Fei-Fei Li, Ehsan Adeli, Euan Ashley: “MIRAGE: The Illusion of Visual Understanding”, 2026

Follow and subscribe

If you stumbled across this article on the web, subscribe to the “misaligned bits” newsletter and follow Misaligned on Medium, LinkedIn, Threads or Mastodon.

You can now also subscribe to this newsletter directly with your email.