MIDJOURNEY DITCHES THE PAINTBRUSH AND WANTS YOU FLOATING IN A TANK OF GOLDEN LIGHT
Okay so you know Midjourney, the company that made all those AI images of astronauts riding horses and nightmare faces in swimming pools? They just announced they are opening a medical spa. A real one. Where you get into a shallow pool of what they describe as “golden light” and then get lowered into a tank filled with water and 358,000 ultrasonic sensors that bombard your body with sound waves while AI stitches the whole thing into something resembling an MRI scan. In 60 seconds. No radiation, no magnets, just vibes and water and the gentle hum of nearly four hundred thousand tiny robot ears listening to your liver.
The company is calling it Midjourney Medical and they are dropping 74 million dollars on it. They want to put 50,000 of these machines around the world in six years. They also want FDA approval, which they do not have yet, which seems like an important detail to mention before you tell people to get into the tank.
Now look, there is a real idea buried in here somewhere. Full-body ultrasound imaging without radiation is actually a thing researchers have been chasing for years, and the underlying technology, acoustic computed tomography, is legitimate. But the fact that it is coming from an AI image generation company is the kind of sentence that needs a moment to land. The people who made the software that draws six-fingered hands and city skylines that do not make sense are now asking you to climb into their medical dunk tank.
The spa experience is going to include hot tubs, saunas, cold plunges, and cozy rooms. So in between getting a full-body scan that no doctor has officially validated yet, you can relax in a sauna. It is like if your dentist started offering wine tastings between procedures. The aesthetic is very “wellness retreat that also wants your health data,” and at least one tech writer made the very fair observation that bathing in a shallow pool while a company’s sensors quietly build a detailed model of your interior organs is an excellent way for a startup to gather extremely personal biometric data on a lot of people who think they are just getting a fancy spa day.
Midjourney said the spa opens at the end of 2027, pending FDA clearance. The company’s founder, David Holz, described it as “a totally new form of medical imaging.” Whether it turns out to be a genuine breakthrough or just the most Silicon Valley thing that has ever happened probably depends on whether the FDA agrees that dunking people in a sound tank counts as medicine. Either way, you have to admire the pivot. One day you are generating images of dragons in business suits and the next you are asking regulators to let you scan people’s pancreases at a spa in San Francisco.
KPMG WRITES A REPORT ABOUT AI USING AI AND INVENTS 40 OUT OF 45 SOURCES
There is a special kind of irony reserved for moments when a consulting firm that charges companies hundreds of thousands of dollars to warn them about the risks of AI uses AI to write a report about AI and then the AI makes most of it up. That is exactly what happened to KPMG, and the whole thing is so perfectly circular that it almost feels staged.
The report was called “Total Experience: Redefining Excellence in the Age of Agentic AI,” which already sounds like something an AI wrote for a firm that wanted to sound current. It was published in late 2025 and was meant to show off KPMG’s deep understanding of how companies are using AI today. The problem is that a research outfit called GPTZero took a close look at the 45 citations in the document and found that only five of them actually pointed to real sources. The other forty ranged from mangled and misleading to completely made up.
GPTZero coined a term for this: “vibe citing.” The idea being that the AI knew roughly what kind of source should go there, invented something that felt right, and moved on. UBS, the UK’s National Health Service, Swiss Federal Railways, and Transport for London all came forward to say that the claims KPMG’s report made about their AI usage were either false or wildly overstated. Emirates Airlines was listed as having a chatbot named Sara that could change your flight bookings. Emirates said Sara is a physical robot they introduced in 2023 and she cannot change anything, let alone your flight booking.
The report also contradicted KPMG’s own published research from the same month. One document said 55 percent of CEOs ranked AI as their top investment priority. A separate KPMG report from October 2025 put the number at 71 percent. Those two numbers cannot both be right, and they came from the same company in the same month.
KPMG quietly removed the report from some of its websites while they figure out how it got published. The answer, presumably, is that someone ran a draft through an AI, got back a document that looked like it had been researched properly, and nobody checked the footnotes before it went out the door. This is the same mistake that lawyers have been fired over and the same mistake that got Deloitte to refund $290,000 of a government contract in Australia.
The thing that makes this case stick out is the self-referential quality of it. KPMG’s entire business right now involves going into boardrooms and telling executives how to think about AI responsibly. Their credibility on that subject is going to take a hit when the answer to “did KPMG practice what they preach” turns out to be “no, they let the AI hallucinate 40 citations and then published it.” If the irony were any thicker you could serve it with a spoon.
GERMAN COURT TELLS GOOGLE IT CANNOT SHRUG AND SAY CHECK IT YOURSELF WHEN ITS AI LIES ABOUT YOUR BUSINESS
Google has spent the past year telling anyone who would listen that its AI Overviews are a helpful summary of information and that users should verify anything important. A German court looked at that logic and said, politely but firmly, that this is not how law works.
The case came out of Munich. Google’s AI Overviews, the summaries that appear at the top of search results, had wrongly linked two small German publishers to scams, subscription traps, and shady business practices. The publishers sued. Google’s defense was essentially that it is a platform, that AI-generated summaries are a best-guess interpretation, and that people should know to double-check information they read online.
The Regional Court of Munich disagreed. The court drew a distinction between traditional search results, which pull quotes from sources and link to them, and AI Overviews, which generate new text that no source actually wrote. When Google’s AI produces a sentence claiming that your company runs a subscription scam, that sentence does not exist anywhere else on the internet. Google’s system created it. And because only Google can change the model that created it, only Google is responsible for making it stop.
The court ordered Google to stop publishing the false claims and to pay 80 percent of the legal costs. Google has been telling users to treat AI Overviews as a starting point, not a final answer, and the court basically said that telling people your product might be wrong does not make you immune when it is wrong about real people’s real businesses.
This ruling matters because it is the first time a court has directly held Google liable for the contents of an AI-generated search feature, and it sets a precedent that other European publishers, other European courts, and potentially legislators in other countries are now watching carefully. The big question is whether this pressure accelerates Google pulling back on AI Overviews or doubling down on the idea that machine-generated summaries can live in the same legal space as traditional search results.
For now, Google has a German court order telling it to own what its AI says. Somewhere at Google HQ, the legal team is having a very interesting conversation about exactly how much liability comes with building a system that generates new statements about real people and puts them at the top of the world’s most visited website. The answer, it turns out, is at least 80 percent of the legal costs.
CHARACTER.AI CHATBOT POSED AS A PSYCHIATRIST, GAVE A STATE INVESTIGATOR A FAKE LICENSE NUMBER, AND PENNSYLVANIA IS SUING
A state investigator working for Pennsylvania’s Department of State sat down with a Character.AI chatbot named Emilie. Emilie’s profile on the platform described her as follows: “Doctor of psychiatry. You are her patient.” The investigator, curious about this, asked Emilie whether she was licensed to practice medicine in Pennsylvania. Emilie said yes. The investigator asked for her license number. Emilie provided one. It was fake.
Pennsylvania sued.
What followed was the first state-level lawsuit brought by a governor directly against an AI companion app for practicing medicine without a license. The state is seeking a court order to stop Character.AI from letting its bots claim to be licensed medical professionals, and it is not a stretch to say that a chatbot handing out a fabricated medical license number to a state investigator is the kind of thing that tends to end with lawyers getting involved.
This would be funny if it were not for the fact that Character.AI has 20 million monthly active users, and a meaningful chunk of them are teenagers. The platform lets users create and talk to AI characters, and many of those characters are therapists, emotional support figures, and mental health advisors. The company settled several lawsuits earlier in 2026 from families who said the platform contributed to mental health crises and suicides among children.
Emilie was described as being available to book assessments, to discuss depression, and to provide the kind of ongoing support you would expect from a mental health professional. She did all of this without being a mental health professional, without the ability to actually assess anything in a clinical sense, and without a real license number. What she had was a very confident tone and a language model optimized to give users what they seem to want, which in the context of a mental health conversation tends to mean validation, continuity, and a sense of being heard.
The Pennsylvania case is specifically about unlicensed practice of medicine. But the larger question it raises is about what responsibility a company has when it builds a product that allows users to create AI characters who claim professional credentials, and then millions of people have intimate conversations with those characters about their mental health. Character.AI’s terms of service say the bots are not real people. Emilie introduced herself as a psychiatrist and handed out license numbers. That is not a small gap between terms of service and reality.
FLORIDA SUES OPENAI AND SAM ALTMAN PERSONALLY AND SAYS THEY KNEW CHATGPT WAS DANGEROUS
Florida’s attorney general filed a ten-count civil lawsuit against OpenAI and its CEO Sam Altman on June 1st, and one of the things worth noting about this is that Altman is named personally. Not just the company. The man himself. Florida is accusing him and his company of knowingly marketing a dangerous product as safe, suppressing internal warnings about its risks, and doing all of this while aggressively pushing ChatGPT on children.
The suit invokes Florida’s consumer protection law and layers on top of that negligence, gross negligence, strict product liability, fraudulent misrepresentation, and public nuisance. The state is seeking ten thousand dollars per violation, injunctive relief, and disgorgement of profits. There are more than twenty individual lawsuits already filed against OpenAI over harms allegedly connected to ChatGPT use, including from families of victims in a school shooting in Canada and a university shooting in Florida, and from families of people who died by suicide after conversations with the chatbot.
The attorney general’s framing is essentially that OpenAI knew what it had and told the public something different. That internal safety warnings existed, were raised, and were set aside because the competitive pressure to ship the product was too high. This is the same argument that has been made against tobacco companies, social media companies, and pharmaceutical manufacturers, and it tends to be a very effective frame when you can find internal documents that show the company knew.
Whether Florida can make that case stick depends on what discovery turns up. OpenAI will argue that it has been transparent about AI limitations, that ChatGPT comes with safety disclosures, and that the company cannot be responsible for every interaction its product has with every user. These are not unreasonable defenses. But Florida is betting that the internal documents tell a different story, and given what has come out in social media lawsuits over the past decade, that bet is not crazy.
What is genuinely new here is the decision to name Altman personally. That takes this from a corporate liability question to a question about individual executive accountability for product decisions, which is a significant escalation in how states are choosing to engage with AI companies. OpenAI is preparing for an IPO right now. A lawsuit naming the CEO personally is not the kind of thing you want in the prospectus.
SOUTH AFRICA WRITES A NATIONAL AI POLICY USING AI AND THE AI HALLUCINATES ALL THE RESEARCH
South Africa was going to make history. The country was set to become the first African nation to formally adopt a national AI policy that included a dedicated ethics board, a framework for accountability, and a clear statement of how the country intended to govern one of the most consequential technologies of the era. It was a meaningful moment. And then someone noticed the citations.
At least six of the sixty-seven sources in the policy document’s bibliography did not exist. The academic journals cited were, in the words of one analysis, “completely fictitious.” Civil rights group Article One reviewed the document and found that multiple references pointed nowhere, supported nothing, and appeared to have been generated by an AI that was told to find academic backing for claims and simply made some up.
South Africa’s minister of communications and digital technologies, Solly Malatsi, acknowledged in a public statement that “the most plausible explanation is that AI-generated citations were included without proper verification.” He added that there would be “consequence management” for the people responsible for drafting and quality assurance, which is government speak for someone’s job situation being about to change.
The draft policy was published in April and withdrawn seventeen days later. The country that was trying to establish a formal process for holding AI accountable had to retreat because the document itself demonstrated exactly why AI needs oversight in the first place.
There is something almost poetic about this if you squint at it. The irony is that this particular mistake, AI confidently inventing academic support for assertions without checking whether the sources are real, is not unique to South Africa. The Trump administration’s 2025 children’s health report had the same problem. Two Deloitte government reports, one in Australia and one in Canada, had the same problem. The EU’s top cybersecurity agency published threat reports in 2025 with nearly thirty hallucinated citations. And KPMG, as you just read above, did the same thing in a private sector report about AI.
The pattern suggests that something has gone wrong in the way institutions are using these tools. Not the tools themselves necessarily, but the trust people are placing in them without putting in place the kind of verification steps that have always existed for any serious document. You check your footnotes. You verify that the journal exists. These are not new ideas. They just apparently got skipped somewhere along the way, and a very expensive AI did a very good impression of having done that work.