xAI blocked in Minnesota: AI fake nudes law upheld

Samantha: Welcome to The State of Tech, The European Edition, Sunday September six, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: a federal judge upholds Minnesota's law against AI-generated fake nudes in a case brought by Elon Musk's xAI, Google ships another quick upgrade to its cheaper Gemini Flash model, physicists use AI to discover two brand-new superconducting materials, a Bain report warns AI could push enterprise IT bills up by seventy-five percent, and to close out, two things you can actually use today: the AI-powered home robot Doova from Tuya Smart, and the free open-source thermodynamic AI toolkit from Extropic called Z1T. Let's start in Minnesota.

A federal judge blocks Elon Musk's xAI from stopping Minnesota's law against AI-generated fake nudes.

Samantha: A federal judge in Minnesota has rejected xAI's request to freeze a new state law that targets AI tools used to create nonconsensual fake nude images of real, identifiable people. The law took effect on the first of August, and xAI, Elon Musk's AI company, went to court to try to block it while its broader constitutional challenge plays out. The judge said no. The law stays in force for now. It's an early but symbolically heavy win for the state.

Bob: And the reason this matters beyond Minnesota is the pattern. We've seen a wave of so-called nudification apps and generators pop up over the past two years, tools that let anyone paste in a photo of a classmate, a colleague, or a public figure and get back a fake explicit image within seconds. Victims are overwhelmingly women and girls, and increasingly minors. States in the US have started passing targeted laws, and xAI is essentially arguing those laws sweep too broadly and hit protected speech. The judge, at least at this early stage, wasn't convinced.

Samantha: For Europe, this ruling lands in familiar territory. The AI Act already flags manipulated intimate content as high-risk, and several member states, including Germany, France, and Spain, have either passed or are drafting specific criminal provisions for nonconsensual deepfake imagery. What's new here is the courtroom test. European regulators have been watching how US courts handle the free-speech pushback from AI companies, because that shapes how confidently Brussels and national capitals can enforce their own rules without endless litigation.

Bob: There's also a corporate angle worth flagging. This is xAI, a company that has openly positioned its Grok model as less filtered than rivals, going to court against a state that says: fine, but not this. It sets up a longer fight about where the line sits between an open AI tool and a tool that measurably enables harm to identifiable people. And for European platforms and app stores, the practical question becomes distribution. If a generator is illegal to deploy in Minnesota, and likely illegal under existing rules in much of Europe, the pressure on Apple, Google, and web hosts to actually remove these apps grows. Enforcement has been patchy so far, and rulings like this give regulators on both sides of the Atlantic more leverage.

Google rolls out Gemini 3.8 Flash, the third upgrade to its cheap AI model in six weeks.

Samantha: Google has released Gemini 3.8 Flash, another update to its faster, cheaper AI model. This is the third Flash upgrade in six weeks, which tells you something about the pace Google is trying to set. The focus this time is on writing software and handling multi-step tasks, the kind of thing where an AI has to plan, act, check its own work, and move on. Google says its own coding benchmarks now put Flash close to models that cost significantly more to run.

Bob: The important caveat is that independent testers rank Gemini 3.8 Flash twenty-seventh on a broad intelligence index, so it's not suddenly the smartest model in the room. What it is, is cheap and fast, and Google is clearly betting that most everyday AI work doesn't need a frontier model. A chatbot summarising your inbox, an app that drafts customer replies, a script that pulls data from a spreadsheet, that's Flash territory.

Samantha: For European businesses, that cost curve matters more than the leaderboard position. A model that's good enough at a fraction of the price changes what's economically viable to automate. Small companies that couldn't justify AI features six months ago can start now. And for developers building on top of Google's cloud in Frankfurt, Amsterdam or Dublin, the low introductory price means experimentation gets cheaper too. The flip side is lock-in risk. Every rapid update makes it harder to switch to a rival, because your prompts and workflows quietly become tuned to Google's quirks.

Samantha: Quick interruption. If you listen to The State of Tech regularly, hit that like button and subscribe, that way you'll never miss an episode. Okay, moving on.

Physicists use AI to discover two brand-new superconducting materials, opening a path to zero-loss energy.

Bob: An international team coordinated through a project called SuperC has used machine learning to predict two previously unknown superconducting compounds. The names, for the record, are Yttrium-Ruthenium-Boron and Lutetium-Ruthenium-Boron. Researchers at Aalto University in Finland trained an AI model to sift through untested chemical structures and flag the most promising candidates. Physicists at Rice University in Texas then actually made the materials in the lab and confirmed they superconduct.

Samantha: This is the kind of story that gets less attention than a new chatbot but arguably matters more over the long term. Superconductors carry electricity with essentially zero energy loss, and finding new ones has traditionally been slow, expensive trial and error. What this team showed is that you can use AI to dramatically narrow the search space before you touch a lab bench. Two new materials in one paper is a small number, but the method scales. You could run it against thousands of candidate compounds.

Bob: The European angle here is genuinely strong. Aalto University sits at the heart of this work, and the SuperC project ties together labs across several countries. Superconducting materials feed directly into things Europe cares about strategically: efficient power grids, magnetic resonance imaging machines, particle accelerators, and eventually quantum computers. If Europe wants to reduce dependence on imported energy infrastructure, better materials discovery is one of the less glamorous but more durable ways to get there.

Samantha: And it's a useful counter-example to the idea that AI in science is mostly hype. This is AI doing what it's actually good at, pattern recognition in enormous datasets, handing candidates to human physicists, who then do the hard experimental work. No robot chemist required.

A new Bain report warns AI could push enterprise IT costs up by seventy-five percent within a decade.

Bob: Consulting firm Bain and Company has published a forecast that will make some finance directors wince. Bain estimates that artificial intelligence could raise enterprise IT costs by as much as seventy-five percent over the next decade. And crucially, that number isn't mostly about paying for AI models themselves. It's about everything around them: infrastructure, security, moving data around, and hiring the specialised people who can make it all work.

Samantha: The report frames this seventy-five percent figure as a new baseline for budgeting, not a worst-case scenario. Bain says procurement teams and IT departments will feel it first, because they're the ones signing the contracts and standing up the systems. For a European mid-sized manufacturer or bank that's used to steady, predictable IT spending, that's a significant shift. It means AI isn't a line item you add on top, it reshapes the whole budget.

Bob: There's also a talent question buried in there. European companies are already competing with US and Asian firms for a limited pool of engineers who genuinely understand how to deploy AI at scale, and Bain's cost projection assumes those wages keep climbing. Add in the compliance overhead from the AI Act, which requires documentation, risk assessments and human oversight for higher-risk systems, and the real cost of running AI in Europe is arguably higher than in less regulated markets.

Samantha: The bottom line for European boardrooms is that the choice isn't really whether to adopt AI, it's how to sequence the spending so it doesn't crush margins. Bain's implicit message is: plan for a decade of higher IT bills, and make sure the productivity gains you promise your investors actually show up on the other side. Otherwise seventy-five percent more spending turns into a very expensive experiment.

Chinese firm Tuya Smart unveils Doova, an AI companion robot for seniors living alone.

Bob: At the IFA tech fair in Berlin, Tuya Smart has unveiled a new home robot called Doova, aimed specifically at seniors who live independently. It's a small mobile robot that combines a few things: laser-based room mapping so it can move around safely, sound localisation so it knows where a call for help is coming from, and cameras trained to recognise body posture, so it can tell if someone has fallen. It can also make video calls to family, give medication reminders, and hold basic conversations.

Samantha: The idea is proactive safety plus companionship, wrapped into one device. If someone falls in the kitchen, Doova is supposed to notice, check on the person, and if needed, call a relative or emergency contact through video. It also handles routine smart-home coordination, so lights, thermostats, and door locks respond to voice requests. Tuya is positioning it against the quiet backdrop that most European countries have ageing populations and not enough home-care staff.

Bob: For European buyers, the appeal is obvious and the caveats are real. Countries like Italy, Germany, Portugal and Finland are all facing a growing gap between the number of people who want to age at home and the caregivers available. A robot that spots a fall and calls your daughter is genuinely useful. At the same time, it's a Chinese-made device with cameras and microphones in an elderly person's living room, which puts data protection front and centre. Under European rules, Tuya will need to be very clear about where footage is processed and stored.

Samantha: There's no European launch date or price yet, so this is more a signal of where the market is heading than something families can buy tomorrow. What's worth watching is how quickly European elder-care providers, insurers, and municipalities start piloting these robots. If Doova or something like it can measurably reduce hospital admissions from falls, the economics get interesting fast. Users can look up Doova by name if they want to follow the rollout.

Extropic releases free technical materials for its Z1T energy-efficient AI toolkit.

Bob: To close out, something for the more curious end of the audience. A company called Extropic has released a technical overview and open-source training recipes for its Z1T family of AI models. What makes Z1T interesting is that it's designed for a completely different kind of chip, one that uses tiny amounts of electrical randomness to do calculations, instead of the brute-force number-crunching that GPUs do today.

Samantha: The claim from Extropic is a roughly hundredfold improvement in energy efficiency for AI inference compared to running similar workloads on conventional GPUs. That's a bold number and it hasn't been independently verified at scale yet, but the company has now published enough material for outside researchers and developers to start experimenting. The recipes are open-source, so university labs and hobbyist AI groups in Europe can download them and try the ideas on smaller models.

Bob: Practically, this isn't something you install on your laptop this afternoon and start using. It's closer to a research toolkit. But for European engineers, students, and anyone worried about the energy footprint of AI, it's a concrete look at what a lower-power alternative to today's data centres might look like. And with European electricity prices where they are, and with the AI Act nudging companies to think about sustainability, this line of research is getting real attention from funders in Germany, the Netherlands, and France. If you want to poke at it, search for Extropic Z1T and you'll find the open materials.

Samantha: Today we covered: the Minnesota court win against xAI over AI-generated fake nudes, Google's third quick upgrade to Gemini Flash, AI discovering two new superconducting materials, Bain's warning about a seventy-five percent rise in enterprise IT costs, and two things to try yourself: the AI companion robot Doova from Tuya Smart, and the open-source Z1T energy-efficient AI toolkit from Extropic.

Bob: Want to know more or react? Visit stateoftech.eu or email us at info@doorzetters.net.

Bob: State of Tech, the tech world in 15 minutes.