AI Safety Body: OpenAI, Google & Anthropic + EU Chips Act 2.0

Samantha: Welcome to The State of Tech, The European Edition, Friday September twenty-fifth, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: OpenAI, Google and Anthropic team up on a new AI safety body; the EU pushes a Chips Act 2.0 to cut its semiconductor dependence; Anthropic's Claude uncovers what could be a new gene editing system; Google prepares to fire AI chips into orbit; and to close out, two things you can actually use today: the free note-taking app Standard Notes Sync, and the open European transport planner Öffi. Let's start with that new safety pact.

OpenAI, Google and Anthropic join forces to launch an independent AI safety standards body.

Samantha: The three biggest names in generative AI are reportedly setting up a joint organisation dedicated to AI safety standards. According to sources briefed on the plans, OpenAI, Google and Anthropic want an independent body that writes safety guidelines, supports third-party testing of AI models, and defines who is qualified to audit them. The target launch is late 2026 or early 2027. It is essentially industry self-regulation, arriving just as governments across the world are still figuring out how to keep pace with these systems.

Bob: And the timing is not accidental. This follows earlier public-private safety pacts, and it lands in a week where reports keep surfacing about AI models being used or misused for hacking activities. The idea is to get ahead of that story with a common rulebook. But critics are already pushing back. A body designed by the three biggest labs risks setting standards that only the biggest labs can meet. Open-source developers and smaller European players fear being locked out of markets by audit requirements they simply cannot afford.

Samantha: That criticism matters, because self-regulation only works if the referee is credible. If the same companies writing the rules also decide who passes the audits, the incentives get uncomfortable fast. Supporters argue that industry knows the technology best and can move faster than any regulator, and that a shared testing standard is better than fifty different national approaches. Both things can be true. The real test will be governance: who sits on the board, who can veto standards, and whether independent researchers get access to the models being audited.

Bob: For Europe, this is a direct nudge to the AI Act machinery in Brussels. European regulators have been building out codes of practice for general-purpose AI models, and a global industry standard could either complement that work or quietly undercut it. If American labs define what counts as a safe model and get third-party auditors certified on their terms, European authorities may find themselves importing a framework rather than shaping one. Expect the Commission and national regulators to want a seat at the table before this body opens its doors.

Samantha: And for European AI companies, especially the open-source community around Mistral and the research labs in Germany and the Netherlands, the worry is concrete. If auditing qualifications are written around closed frontier models, an open-weight model released for free could still need an expensive certification to be deployed commercially. Industry groups are likely to argue for tiered rules that do not treat a small research release the same as a hundred-billion-parameter commercial system.

EU ministers push a Chips Act 2.0 to cut Europe's dependence on foreign semiconductors.

Bob: European Union industry ministers are actively discussing what is being called a European Chips Act 2.0, aimed at significantly reducing the bloc's reliance on foreign semiconductor suppliers. The new plan tries to move beyond simply subsidising chip factories. Instead, it wants to stimulate demand, strengthen supply chains, and build resilience against future disruptions. It is a direct response to the shortages that hit European carmakers and medical equipment makers hard in recent years, when a missing chip could halt an entire assembly line.

Samantha: The shift in thinking here is important. The original Chips Act poured money into building capacity, but Europe still ended up dependent on the same handful of Asian and American suppliers for the most advanced components. This version leans on public procurement and market access to link European chipmakers directly to European industrial buyers. In plain terms, if Brussels or a national government is buying medical scanners or defence equipment, the chips inside should preferably come from European fabs. That is a much more industrial-policy approach than pure subsidy.

Bob: Critics will point out that Europe still lacks the leading-edge manufacturing that TSMC in Taiwan or Samsung in South Korea offer, and no policy discussion in Brussels changes that overnight. What this act can do is secure the specialist chips European industry actually uses day to day: automotive controllers, power semiconductors, sensors. Not the flashy AI accelerators, but the boring components that shut down a Volkswagen plant when they run out.

Samantha: For European businesses reliant on chips, this could translate into more stable supply and possibly better pricing on domestically produced components, though not immediately. The bigger question is whether member states will actually align on procurement rules, because national interests around Intel in Germany, STMicroelectronics in France and Italy, and ASML in the Netherlands rarely point in the same direction. If the ministers can agree on a coherent demand-side strategy, Europe genuinely improves its technological sovereignty. If they cannot, this becomes another well-intentioned framework that changes little on the ground.

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.

Anthropic says its Claude AI autonomously discovered a possible new CRISPR-like gene editing system.

Bob: Anthropic is claiming a genuinely striking scientific result. Its flagship model Claude, running for twenty-one hours through a vast database of DNA sequences, identified a novel enzyme system inside bacterial DNA that shows characteristics similar to CRISPR, the gene editing tool that reshaped biology over the past decade. Researchers at Anthropic's biology lab set up the search, and Claude flagged the candidate autonomously. CEO Dario Amodei suggests this could represent a new gene editing mechanism, though the precise function is still being investigated.

Samantha: The important framing here is that Claude did not invent biology. It sifted through enormous quantities of existing sequence data and spotted a pattern that fits the profile of a gene editing enzyme. Human researchers still need to validate it in the lab, test whether it actually cuts DNA the way CRISPR does, and figure out if it offers any advantage over existing tools. So this is a lead, not a finished discovery. But it is the kind of lead that could take a human research team months to find, compressed into under a day.

Bob: If it holds up, it is another data point in a growing pattern. AI models are becoming genuinely useful for finding needles in scientific haystacks, whether that is protein structures with DeepMind's earlier work, new antibiotic candidates, or now potentially new gene editing tools. What Anthropic is really showing is that a general-purpose model, pointed at a specialised database with the right prompting, can do work that once required dedicated bioinformatics pipelines.

Samantha: For European biotech, this cuts two ways. Research institutes in places like Heidelberg, Cambridge and Leuven get a powerful new way to accelerate discovery, and European pharmaceutical companies could licence or develop therapies based on new gene editing mechanisms. But it also raises familiar questions about who owns discoveries made by an AI system trained on public data. If Claude finds it, who patents it? European regulators and research councils will need to think about that faster than they might like.

Google prepares to launch its first AI chips into orbit under Project Suncatcher.

Bob: Google is preparing to launch a prototype satellite next week as part of what it calls Project Suncatcher. This is a research project exploring whether large-scale AI computing can eventually run in space. The mission is a collaboration with Planet Labs, and it will fly on SpaceX's Transporter-18 rideshare. On board are Google's own AI chips, the Tensor Processing Units it uses to train and run its models. The point of this launch is not to do useful AI work in orbit, but to see whether the hardware survives.

Samantha: And survival is the real question. Launch itself shakes hardware violently, low Earth orbit exposes chips to radiation that flips bits and degrades circuits, and temperature swings between sunlight and shadow are brutal. Consumer-grade silicon simply does not last there without shielding. So Google is stress-testing whether its AI accelerators can be adapted for orbital use at all. Other players are chasing the same idea. SpaceX and a startup called Starcloud are exploring space-based data centres too, betting on continuous solar power and the fact that terrestrial grids are struggling to keep up with AI's electricity appetite.

Bob: The pitch is genuinely interesting. In orbit, you get near-constant sunlight for energy and the vacuum of space for cooling, two of the biggest headaches for AI data centres on the ground. Whether it ever becomes economically viable is another question. Launch costs are falling but still enormous, and repairing a broken server in orbit is not exactly a service call.

Samantha: For European space agencies and the growing commercial space sector on the continent, this is a signal to watch closely. The European Space Agency, Airbus Defence and Space, and a wave of smaller launch startups are all thinking about what comes after telecommunications and Earth observation. If space-based compute becomes real, Europe will want a strategic position in it, both for sovereignty reasons and because energy-hungry AI infrastructure is becoming a political issue in countries where data centres compete with households for grid capacity.

The encrypted note-taking app Standard Notes Sync keeps years of writing genuinely private across every device.

Bob: To close out, two things you can actually use today. The first is Standard Notes Sync, a note-taking app that has quietly built a reputation for taking privacy seriously. It uses end-to-end encryption by default, meaning even the company running the servers cannot read what you write. It works on Windows, Mac, Linux, iPhone, Android, and in any web browser, and there is a free tier that covers plain text notes across all your devices. A paid tier unlocks rich text, code editors, spreadsheets, and file attachments.

Samantha: What makes it interesting compared to the usual suspects is the focus on longevity. The developers publish the encryption specification openly, so if the company ever disappeared, your notes are stored in a format that can still be decrypted with your password. That is a rare commitment in an industry where cloud services shut down and take your data with them. Users often mention it in the same breath as journalists, lawyers and researchers who need to know that sensitive drafts are not being scanned to train someone else's AI model.

Bob: The setup takes about ten minutes. You create an account, pick a strong passphrase, and start writing. The important detail is that the passphrase is the encryption key, so if you forget it, nobody at the company can help you recover your notes. That is the trade-off with true end-to-end encryption. Write it down somewhere safe, or store it in a password manager.

Samantha: For European listeners, the app is popular partly because it aligns well with what GDPR asks companies to do around data minimisation. The company itself is based outside the EU, but the encryption model means jurisdiction matters less than it usually would. If you have been meaning to move your journal, meeting notes or research out of Google Keep or Apple Notes, this is a serious afternoon project.

Standard Notes Sync, private encrypted notes across all your devices.

The open European transport planner Öffi navigates public transport across Europe without tracking.

Bob: The second pick is Öffi, an Android app that has become a quiet favourite among people who travel by public transport across Europe. It plans journeys, shows live departures, and gives you real-time delay information for buses, trains, trams and metros across Germany, Austria, Switzerland, the Netherlands, Belgium, Italy, Poland, Ireland, and more. It works with the official transit data from each country's operators, so accuracy is genuinely good.

Samantha: What makes it stand out is what it does not do. There are no advertisements, no accounts to create, no location tracking beyond what you need for a route lookup, and no telemetry sent back to anyone. It is free, open source, and available through F-Droid, which is the alternative app store for open-source Android software. There is also a version on the Google Play Store for people who prefer that route.

Bob: Reviewers regularly praise how quickly it launches and how little battery it uses compared to the big commercial apps. If you commute daily, that adds up. And unlike Google Maps or Citymapper, it does not fall back on your search history to personalise results, which some users appreciate and others find slightly less convenient. It is a fair trade if privacy matters to you.

Samantha: The main limitation is that it is Android only, so iPhone users will need to look elsewhere. But if you are travelling by train from Berlin to Vienna next month, or navigating trams in Amsterdam, it is worth a download this afternoon. Today we covered: the new AI safety body from OpenAI, Google and Anthropic; the EU's Chips Act 2.0; Anthropic's Claude finding a possible new gene editing system; Google launching AI chips into orbit; and two things to try yourself: Standard Notes Sync and Öffi.

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.