GPT-5.6 public release: OpenAI, safety review & AI Act

Samantha: Welcome to The State of Tech, The European Edition, Tuesday July fourteen, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: OpenAI opens up its most powerful models yet to the public, China cracks down on AI romance apps, Utah lets a chatbot refill your prescriptions, a shape-shifting liquid metal robot straight out of science fiction, and two things you can actually try today, the free writing tool LanguageTool and the open-source photo app Ansel. Let's start with OpenAI.

OpenAI opens its most advanced GPT-5.6 models to the public after government safety review.

Samantha: OpenAI is about to make its most powerful AI models yet available to everyone. The company confirmed that its new GPT-5.6 family, three models named Sol, Terra and Luna, will go public later this week. The flagship, Sol, is tuned for heavy-duty reasoning, coding and scientific work, and OpenAI claims state-of-the-art results across all three areas. What's notable is that this launch follows a formal safety review carried out together with government partners, a first for a release of this scale.

Bob: And that government-in-the-loop part is the real story here. OpenAI is essentially saying: yes, we ran our most capable model past official assessors before shipping it, but no, we don't want that to become the default for every future release. That's a careful line to walk. On one side, regulators and safety researchers have been pushing for exactly this kind of pre-release check. On the other, OpenAI is signalling it doesn't want a permanent handbrake on how fast it can ship. The tension between speed and oversight is now out in the open, with a real product attached to it.

Samantha: For the European market, this is more than a product launch. European developers, startups and researchers get access to a model class that until recently was reserved for closed pilots. That matters because a lot of European companies have been building on GPT-4-era models while quietly waiting for the next big jump. Sol in particular, aimed at scientific and coding work, lands right in the middle of Europe's push to build sovereign AI capacity in research, pharma and engineering.

Bob: There is a regulatory subtext too. Europe's AI Act obligations for the most capable general-purpose models are kicking in, and Brussels has been asking for detailed documentation, risk assessments and post-market monitoring. A US-led government safety review is not the same as compliance with European rules, but it does create a template. If OpenAI can point to an American assessment and then map that onto EU requirements, deployment in Europe becomes faster. Expect European cloud partners and enterprise buyers to move quickly once the models go live, because the queue for high-end reasoning capacity is already long.

China moves to shut down AI romance companions over dependency and exploitation fears.

Samantha: China is stepping in on one of the fastest-growing corners of consumer AI: emotionally intimate chatbots. Beijing is preparing new rules that take effect next week specifically targeting anthropomorphic AI romance companions. Regulators say these apps risk exploiting loneliness, deepening dependency and turning intimacy into a commercial product, with particular concern about the impact on women. ByteDance, Alibaba and Tencent have already started pulling features that let users create personalised AI partners, ahead of the deadline.

Bob: This is a striking regulatory move because it targets emotional design, not just data protection or misinformation. Companion chatbots have quietly become a huge market in China, with millions of paying users. The rules focus on how these systems shape behaviour: the memory features that make an AI feel like a long-term partner, the paid gifts, the escalating emotional attachment. Beijing is basically saying that engineering intimacy at scale is a public health issue.

Samantha: European regulators will be watching closely, because the same products exist here, and the same debates are starting. Character-style apps, AI girlfriends and boyfriends, virtual partners built on top of large language models, all of it is available on European app stores today, largely unregulated at the product level. The AI Act touches manipulation and vulnerability, but it does not spell out rules for emotional dependency the way China now will. If Beijing's approach shows measurable public health benefits, expect that evidence to feed directly into European policy debates over the next year.

Bob: And there's a competitive angle. Chinese tech firms losing a lucrative product category at home might push those same features harder in export markets, including Europe. That could mean European users end up as the testing ground for exactly the products Chinese users are being protected from. Consumer groups and data protection authorities across the continent will likely take notice.

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.

Utah lets an AI chatbot refill prescriptions without a doctor, igniting a national medical debate.

Bob: Utah has just done something no US state has tried before. It's launched a pilot programme letting residents refill prescriptions online through an AI chatbot called Doctronic, with no traditional doctor visit in the loop. The programme runs under what's known as a regulatory sandbox, which allows state officials to waive existing rules for AI technologies they consider promising. Doctors, lawyers and public health experts are, unsurprisingly, split down the middle.

Samantha: The debate is real. On one side, refilling a stable prescription is one of the most routine tasks in primary care, and automating it could free up doctors for genuinely complex cases. On the other, even a refill involves judgment: is the medication still working, are there new side effects, has something else in the patient's life changed. Handing that to a chatbot, even a well-designed one, means trusting the AI to know when to escalate to a human. And the sandbox model raises a bigger question: how far are regulators willing to bend existing licensing laws for AI systems that have not been through the usual clinical trials.

Bob: For Europe, this is a preview of a conversation that's coming. The European Medicines Agency and national health regulators have been cautious about AI in prescribing, and medical licensing rules make a Utah-style waiver essentially impossible in most member states today. But the pressure is building. General practitioners across Europe are overloaded, waiting times are climbing, and digital health startups are lobbying hard for lighter regulation on repeat prescriptions and low-risk medication management.

Samantha: The key thing to watch is what happens in Utah over the next twelve months. If patient outcomes hold up and there are no serious safety incidents, expect European health ministries to face renewed pressure to allow similar pilots, probably starting in the Nordic countries and Estonia, where digital health infrastructure is most advanced. If it goes badly, it becomes a cautionary tale that slows AI in medicine across the entire continent.

Korean scientists build a magnetic liquid metal robot that splits, merges and squeezes through gaps.

Bob: Now this one genuinely reads like science fiction. Researchers at Seoul National University and Gachon University have built a magnetic liquid metal robot that can change shape on command. It splits into multiple droplets, merges back into a single body, and squeezes through gaps far smaller than itself. They call it a particle-armoured liquid robot, and it takes inspiration from how biological cells behave, combining the flexibility of a liquid with the structure of a solid.

Samantha: The clever engineering here is the armour part. Pure liquid metal on its own is unstable and messy. By coating it with magnetic particles, the researchers get something that holds its shape when it needs to, deforms when required, and responds precisely to external magnetic fields. That combination has been the holy grail of soft robotics for years. The potential applications are exactly where rigid robots struggle: minimally invasive surgery inside the human body, inspecting the inside of complex machinery without disassembly, and disaster response in collapsed buildings where rescuers cannot reach.

Bob: Europe has strong soft robotics research, particularly in the Netherlands, Germany, Italy and Switzerland, and this kind of breakthrough tends to travel fast through the academic world. Expect European labs to be replicating and extending the work within months. The medical applications are especially relevant, because Europe has a large medical device industry and a regulatory framework, the Medical Device Regulation, that is already used to evaluating novel robotics. That gives European hospitals and manufacturers a realistic path to trialling this technology once it matures.

Samantha: The catch is time. Lab demonstrations are not clinical products, and turning a shape-shifting droplet into something safe enough for a surgical suite or an industrial pipeline will take years. But it is a genuine step forward, and it moves soft robotics out of the "interesting idea" category and into "credible engineering path".

LanguageTool is the free European grammar checker that runs privately and speaks thirty languages.

Bob: Time for two things you can actually try today, starting with something for anyone who writes for a living, or just writes emails they don't want to embarrass themselves with. LanguageTool is a grammar and style checker built by a German team, and it has quietly become one of the strongest alternatives to the big American writing assistants. It works in over thirty languages, including all the major European ones, and it catches things a plain spellchecker misses: awkward phrasing, tone mismatches, false friends between languages, and grammar patterns specific to non-native speakers.

Samantha: What makes it stand out for European users is where the processing happens. There's a free browser extension for Chrome, Firefox and Edge that works across Gmail, Outlook, Google Docs, LinkedIn, basically anywhere you type. There are also add-ins for Microsoft Word and Google Docs, and a standalone desktop app for people who prefer not to run browser extensions. The free tier handles most everyday writing. A paid Premium tier adds deeper style analysis and longer text checks, but many users say the free version already covers most needs.

Bob: The privacy angle is the real selling point. LanguageTool is based in Germany, subject to European data protection rules, and offers a self-hosted version for organisations that want text checking without any data leaving their own servers. Law firms, hospitals and government agencies across Europe have adopted the self-hosted option for exactly that reason. For an individual user, the free browser extension is a five-minute install, works immediately, and handles multilingual writing better than most competitors. Reviewers particularly praise it for catching the kind of subtle errors that native speakers make when switching between English, German, French, Spanish and Dutch.

Samantha: One practical tip: after installing the extension, take a minute to set your mother tongue and the languages you write in. The suggestions get noticeably sharper once it knows which false friends and grammar patterns to look for. That small setup step is where a lot of users get more value out of it than they expect.

Ansel is the free open-source photo editor giving Lightroom a serious European alternative.

Bob: And to close, something for anyone with a camera roll that has quietly grown out of control. Ansel is a free, open-source photo editor built by a small European team as a friendlier fork of the well-known Darktable project. It handles raw files from every major camera brand, offers professional-grade editing tools, and unlike the big subscription products, it lives entirely on your own computer. No cloud upload, no monthly fee, no account required.

Samantha: The pitch is straightforward. Users who have tried it report that Ansel keeps the serious editing power of Darktable but strips out the intimidating complexity that puts casual photographers off. The interface is cleaner, the defaults are more sensible, and the learning curve is much gentler. You still get proper raw processing, colour grading, lens corrections, noise reduction, and non-destructive editing, meaning your original files are never touched. It runs on Windows, Mac and Linux.

Bob: For European users specifically, it hits a sweet spot. Many photographers here have grown wary of subscription-based editing software where your library effectively lives in a US cloud. Ansel keeps everything local, and because it is open source, the code is fully auditable. Photography clubs, journalism schools and freelance photographers across Europe have been recommending it for exactly this reason. It is not a five-minute tool the way LanguageTool is; expect to spend an evening learning where the main controls are. But once set up, it replaces a chunk of expensive software with nothing at all.

Samantha: One useful starting point: Ansel comes with sensible default presets, and reviewers suggest starting with those before diving into manual adjustments. That way you get a feel for what the tool does before wrestling with every slider yourself.

Samantha: Today we covered: OpenAI's GPT-5.6 models going public, China cracking down on AI romance companions, Utah's chatbot prescription pilot, Korea's shape-shifting liquid metal robot, and two things to try yourself, the free grammar checker LanguageTool and the open-source photo editor Ansel.

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.