UN AI Safety Accord: 150+ Nations Sign at Summit

Samantha: Welcome to The State of Tech, The European Edition, Friday July thirty-first, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: a landmark global AI safety deal signed at the United Nations, DeepMind teams up with CERN to speed up particle physics, Sanofi shows early results from an AI-designed cancer drug, a new global standard for quantum-proof encryption, and to close out, two things you can actually try today: the StarNet satellite internet sign-up, and the Ay-Double-You-Es-published guide to home-server liquid cooling. Let's start with the UN accord.

More than one hundred fifty nations sign a landmark global AI safety accord at the UN.

Samantha: This is easily the biggest story of the day. Leaders from more than one hundred fifty countries gathered at a special United Nations summit and unanimously endorsed a new global AI Safety Accord. It sets out a shared framework for how advanced AI should be developed and deployed, with transparency, accountability, and preventing misuse at its core. And critically, it creates a new UN-backed body to monitor compliance and update the rules as the technology moves.

Bob: And that last bit is what makes this different from the usual talking-shop declarations. There is money attached to independent safety research, and there are shared principles specifically aimed at what the accord calls existential risks from advanced systems. That language would have been considered fringe two years ago, now it is in a document signed by heads of state. The accord also pushes for international cooperation on red-teaming, on incident reporting, and on common evaluation standards. So it is not just a values statement, it is the beginning of an infrastructure.

Samantha: For Europe the timing is interesting. The AI Act is already in force here, and Brussels has been arguing for years that AI needs global rules, not just regional ones. This accord validates that approach. It also raises the pressure on companies operating across borders: if the UN body starts issuing guidance that clashes with, say, national rules in the United States, European regulators will likely align with the multilateral track first. Industry groups are already asking how compliance will actually be measured.

Bob: The other question is enforcement. A UN body can monitor and it can name and shame, but it cannot fine a company in San Francisco or shut down a lab in Shenzhen. Critics point out that without teeth, the accord risks becoming symbolic. Supporters counter that even a shared vocabulary and a shared reporting system is a huge step, because right now every country defines AI risk differently. Expect the first concrete guidelines from the new body within twelve months, and watch which frontier labs sign on voluntarily first. That will tell you whether this has real weight or not.

DeepMind and CERN sign a multi-year deal to speed up particle physics with AI.

Samantha: Google's DeepMind and CERN, the European particle physics lab near Geneva, have announced a multi-year partnership to accelerate physics simulations using AI. The idea is to build deep learning models that can chew through the enormous datasets coming out of the Large Hadron Collider much faster than traditional methods. Simulations that currently take weeks on supercomputers could shrink to hours.

Bob: And that matters because CERN is drowning in data. The Large Hadron Collider produces around a petabyte a second at peak, and only a tiny fraction can actually be stored and analysed. If AI models can pre-filter and simulate collisions faster, researchers can test more theories and get to answers sooner. DeepMind has form here, they have done similar work on protein folding with AlphaFold, and that reshaped biology within a few years.

Samantha: What is notable is the direction of travel. A few years ago, big tech AI labs were mostly chasing consumer products and language models. Now they are increasingly plugging into fundamental science, and specifically into European institutions. CERN keeps ownership of the data and the physics, DeepMind brings the compute and the model expertise. It is a template other European research centres are watching closely, because it lets them access frontier AI without handing over the underlying science.

Bob: For the wider European research community this is a boost, but also a reminder. The compute infrastructure is still coming from a US company. The EU has been trying to build sovereign AI capacity for exactly this reason, so that partnerships like this become a choice rather than a necessity. Meanwhile the science benefits immediately, and that is the point. If this leads to genuine discoveries about dark matter or the Higgs sector, it will be a case study in AI as a scientific instrument, not just a chatbot engine.

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.

Sanofi's AI-designed cancer drug shows promising results in first human trials.

Bob: French pharma giant Sanofi has announced encouraging early clinical trial results for a cancer drug that was designed almost entirely by its in-house AI discovery platform. The candidate targets an aggressive form of cancer, and in Phase One trials it showed significant efficacy along with a favourable safety profile. Sanofi says it will fast-track further development.

Samantha: This is one of the first times a big established pharma company, rather than an AI-native biotech startup, has taken an AI-designed compound this far into humans and reported positive results. The industry has been promising this moment for about five years. AI shortens the discovery phase from years to months by predicting which molecules will bind to which targets. But until now, most of those candidates have either failed in trials or never made it out of the lab.

Bob: A Phase One result is still early, it mainly proves safety and hints at efficacy. But if Sanofi can carry this through Phase Two and Three, it changes the economics of drug discovery. Traditional pipelines cost billions and take a decade. If AI can consistently shave years off that, European pharma suddenly has a real answer to American biotech dominance. Companies like Novo Nordisk, Roche and GSK are all running similar programmes and watching this trial closely.

Samantha: For European patients, the bigger picture is that AI-designed drugs may reach clinics faster, but they still have to go through the same regulators. The European Medicines Agency has been working on guidelines for AI-derived candidates, focusing on how to validate the underlying models. Expect that conversation to accelerate now, because a positive Sanofi trial makes this concrete, not theoretical.

The ITU approves the world's first global standard for quantum-safe encryption.

Bob: The International Telecommunication Union has formally approved a new global standard for quantum encryption. It covers both quantum key distribution and post-quantum cryptography algorithms, and it is designed to make sure different countries and vendors can talk to each other securely once quantum computers become powerful enough to break today's encryption.

Samantha: And that threat is closer than most people realise. Security agencies have been warning for years about what they call harvest-now-decrypt-later attacks, where adversaries record encrypted traffic today and simply wait until quantum machines can crack it in five or ten years. Diplomatic cables, medical records, banking data, it is all at risk. Having a shared standard means governments and companies can start rolling out quantum-resistant systems without worrying they will end up incompatible with everyone else.

Bob: European telecoms and banks have been piloting quantum key distribution for a while, mostly in the Netherlands, Germany and Switzerland. This ITU approval gives them a clear framework to move from pilot to production. The EU has also been funding a quantum communications backbone called EuroQCI, and that project can now align with an internationally recognised standard rather than inventing its own.

Samantha: The catch is cost and complexity. Rolling out post-quantum cryptography across banking systems, government networks, and cloud infrastructure is a multi-year job. And organisations that delay risk being caught out. Security researchers point out that migration typically takes five to seven years, so anyone starting today is already cutting it fine. This standard removes one excuse, which was interoperability. The clock is now ticking louder.

StarNet launches its first thirty satellites for global high-speed internet, sign-ups open in Europe.

Bob: To close out, two things you can actually use today. First, a new satellite internet provider called StarNet has launched its first thirty satellites, and importantly for listeners, sign-ups for early access are open in several European countries. The plan is more than five hundred satellites within three years, using laser inter-satellite links to deliver low-latency broadband to rural and remote areas.

Samantha: For anyone living in a village where fibre still has not arrived, or in a mountain valley where mobile coverage drops out, this is worth checking. StarNet says early-access users in parts of Scandinavia, the Alps, Scotland and rural Spain can join a waitlist through the company's website. Speeds are advertised as comparable to terrestrial fibre once the constellation is denser. The initial dish and installation pricing is being positioned below existing satellite competitors, though full commercial pricing is still to come.

Bob: Reviewers note two things worth knowing. One, laser links between satellites mean the network can route around outages, which historically has been the weak point of satellite broadband. Two, StarNet says it will offer a European data-routing option, so traffic stays inside the continent where possible. That is a real differentiator against the American incumbent. If you are curious, the sign-up takes about two minutes and does not commit you to a purchase. It just puts you on the list for when service goes live in your postcode.

Samantha: And if service does not reach your area soon, it is still useful to register interest, because deployment priority is partly based on demand density. Rural community groups in Ireland and Portugal have already coordinated group sign-ups for this reason.

AWS publishes a free technical guide on liquid cooling that home-lab builders can actually use.

Bob: The second thing to try today is a bit different. Ay-Double-You-Es, Amazon's cloud arm, unveiled a new liquid cooling system for its data centres, and alongside the announcement they published a surprisingly detailed technical guide on how dielectric fluid cooling works, including diagrams and efficiency figures. It is free to read on the Ay-Double-You-Es site.

Samantha: Now this is not a consumer product, but for anyone running a home server, a gaming rig, or a small AI workstation, the guide is genuinely useful. It explains why immersion cooling and direct-to-chip liquid cooling deliver up to thirty percent energy savings, and it walks through the trade-offs. Enthusiast forums have already flagged it as one of the clearest public explainers on the topic. If your home lab sounds like a hairdryer in summer, this is background reading that might change how you build your next machine.

Bob: It is also a useful primer if you work in IT and your company is being asked about data centre energy use. The guide is written for engineers but stays readable. It takes maybe twenty minutes to work through, and it is available in English on the Ay-Double-You-Es website with no login required.

Samantha: A rare case of a big cloud announcement that leaves something practical behind for people outside the industry. And with data centre energy consumption rising fast because of AI training, this kind of knowledge is filtering down from hyperscalers to everyday builders faster than it used to.

Samantha: Today we covered: the landmark UN AI safety accord, DeepMind and CERN teaming up on particle physics, Sanofi's AI-designed cancer drug showing early promise, the new global standard for quantum-safe encryption, and two things to try yourself: the StarNet satellite internet sign-up, and the Ay-Double-You-Es liquid cooling technical guide.

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.