US senators move to ban superintelligent AI, Palo Alto's AI cyber defence

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

Bob: And I'm Bob Russell. Today: US senators try to ban superintelligent AI outright, Akamai lands an eleven-billion-dollar cloud deal with Anthropic, Google launches a new AI laptop called the Googlebook, Palo Alto Networks unveils a defence system against AI-driven cyberattacks, and to close out, two things you can actually use today: the free open-source flashcard tool Mnemosyne, and the privacy-friendly weather app Breezy Weather. Let's start with Washington.

US senators propose banning superintelligent AI and jailing violators for up to twenty years.

Samantha: Senator Bernie Sanders and Representative Greg Casar have introduced what they call the Ban Artificial Superintelligence Act. The bill would permanently prohibit AI systems that outperform humans across most cognitive tasks, pause advanced AI development until federal safety rules are in place, and create a cabinet-level Department of Artificial Intelligence. And the penalties are eye-watering. We are talking a corporate death penalty for violators, or up to twenty years in prison, comparable to unlawful nuclear weapons development. That framing alone tells you how serious the sponsors see this.

Bob: The nuclear comparison is deliberate. Sanders and Casar are treating frontier AI as a category of technology that, once released, cannot be recalled. The bill does not just target companies that reach superintelligence, it wants a full pause on the run-up to it until Washington writes the rulebook. A new cabinet department would then oversee licensing, audits, and enforcement. Whether this bill passes in anything like its current form is another matter. It sits well to the left of where most of Congress is on tech regulation right now, and the White House has consistently favoured a lighter touch.

Samantha: The industry response will be worth watching. OpenAI, Anthropic and Google DeepMind have all publicly said they want some form of frontier oversight, but a hard prohibition with prison terms is not what they had in mind. Expect heavy lobbying, and expect the debate to shift toward compromise language around evaluations and reporting rather than outright bans. Still, the bill itself moves the Overton window. Once senators are talking about banning a whole class of AI, softer proposals suddenly look moderate.

Bob: For Europe this lands at an interesting moment. Brussels is still working through implementation of the AI Act, and member states are debating how strictly to enforce the rules on general-purpose models. A serious US proposal to prohibit superintelligence gives European regulators political cover to argue that their own approach is measured, not extreme. It also raises a practical question: if Washington ever did move toward a licensing regime, would European labs face the same restrictions when serving American customers? That is the kind of extraterritorial reach we saw with export controls on chips.

Samantha: And there is a message for European citizens too. The idea that a democratic legislature could simply say no to a certain class of technology, and back it with criminal penalties, cuts against the assumption that AI progress is inevitable. Whether you agree with the bill or not, it reopens a debate many in the industry had quietly closed.

Akamai signs an eleven-point-six-billion-dollar cloud deal with Anthropic, potentially rising to twenty billion.

Bob: Akamai Technologies has announced a seven-year cloud computing agreement with Anthropic worth eleven-point-six billion dollars, with room to grow to twenty billion. The deal focuses on expanding Akamai's distributed infrastructure to handle the compute demands of Anthropic's Claude models. What makes this notable is that Akamai is best known for content delivery, not AI training. This is a company repositioning itself as serious AI infrastructure, and Anthropic is the anchor tenant that makes that pivot credible.

Samantha: And this fits a wider pattern. Anthropic has already lined up major deals with Amazon and Google, and now Akamai. The company clearly does not want to depend on a single cloud provider, which is smart both commercially and strategically. It also confirms just how much raw compute frontier AI now consumes. Twenty billion dollars over seven years, for one customer, at one infrastructure provider, and that is only part of Anthropic's total spend on servers and chips.

Bob: For Akamai the appeal is obvious. Their distributed edge network, originally built to speed up websites, turns out to be useful for running AI closer to users. That matters for latency-sensitive applications like voice assistants and real-time coding tools. Instead of every query going back to a central data centre, inference can happen at the edge.

Samantha: For European businesses and regulators the deal signals continued concentration of AI capacity in the hands of a few American providers. European cloud players like OVHcloud and Deutsche Telekom's T-Systems are working on their own AI-ready infrastructure, but the scale gap is widening. Every multi-billion-dollar deal like this one makes European sovereign cloud ambitions harder to catch up on. It also means European enterprises using Claude will increasingly rely on Akamai's edge network, which does at least have significant presence inside the EU.

Bob: Quick interruption from me. 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. Actually, Samantha, that is your line.

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.

The Vatican Enters the AI Arms Race — With a Prayer and a Peace Plan

Bob: So here's a sentence I genuinely did not expect to say on this show — the Pope is at UNESCO today, talking about AI governance, and honestly? It might be one of the more interesting diplomatic moves of the year.

Samantha: It really is a fascinating one. This is Pope Leo the Fourteenth, and he's giving a major address at UNESCO headquarters in Paris — the first papal visit there in over forty years, by the way — and the core message is: we need international safeguards on AI, and we need them now.

Bob: And the timing is very deliberate. We're in this period of real tension between the US and China over who controls the future of AI, and the Vatican is essentially raising its hand and saying — we'll be the neutral party here. We'll be the moral facilitators.

Samantha: Which sounds lofty, but there's actually a pretty reasonable argument for it. The Vatican has no economic stake in AI dominance. It's not selling chips or foundation models. It has relationships with governments on every continent. If you want a broker that neither Washington nor Beijing can accuse of playing favourites, a two-thousand-year-old institution with global reach is not the worst candidate.

Bob: The framing from the Vatican's side is very much around human dignity and the common good — classic Catholic social teaching, applied to large language models, which is a sentence that would have sounded absurd five years ago.

Samantha: The specific ask is international safety measures — though the speech hasn't laid out a detailed technical framework. It's more of a normative push. A "this is the direction humanity should go" moment rather than a regulatory proposal with teeth.

Bob: Still, the symbolism matters. When a sitting Pope puts AI ethics at the centre of his pontificate this early, it signals to Catholic-majority countries — many of them in Europe, Latin America, parts of Africa — that this is a serious moral issue, not just a tech policy wonk debate.

Samantha: And frankly, the more voices pulling AI governance toward a human-centred framework, the more pressure there is on the big players to actually engage with that conversation rather than race past it.

Palo Alto Networks launches an AI defence system built to fight AI-generated cyberattacks.

Bob: Palo Alto Networks has introduced a new AI Agent System designed specifically to defend enterprises against AI-generated cyberattacks. The idea is that attackers are now using AI to write phishing emails, generate malware variants and probe for weaknesses at machine speed, so defenders need AI moving at the same pace. The system offers visibility into how AI is being used inside an organisation, checks whether real-time activity matches company policy, and is built for quick deployment.

Samantha: The threat picture behind this launch is real. Security researchers have been warning for the past year that AI-generated phishing is now good enough to fool trained employees, and that criminal groups are using large language models to speed up every stage of an attack. Traditional security tools, which rely on known signatures and slower analyst review, cannot keep up. Palo Alto is essentially arguing that only AI can spot AI at that speed.

Bob: There is also a governance angle. The system does not just watch for external threats, it monitors internal AI use. That matters because employees increasingly plug company data into consumer AI tools without permission, what security teams call shadow AI. A platform that flags when someone pastes a customer database into a chatbot is arguably as valuable as one that blocks external attacks.

Samantha: For European enterprises the launch lands on top of tightening rules. The NIS 2 directive requires stronger cybersecurity measures for critical sectors, and the AI Act adds obligations around monitoring AI systems. A tool that ties both together, threat defence and AI governance, will find willing customers. The catch is cost and complexity. Palo Alto's enterprise pricing is not aimed at small businesses, so the mid-market will still be looking for lighter alternatives.

Bob: The wider signal is that cybersecurity budgets are shifting fast. Money that would have gone to traditional firewalls last year is being redirected to AI-aware defence, and that reshuffling will affect every European CISO's planning for 2027.

The free open-source flashcard app Mnemosyne turns any laptop into a serious study machine.

Samantha: To close out, two things you can actually use today. First one is Mnemosyne, a free open-source flashcard program that has been quietly refined over more than a decade. It uses spaced repetition, the technique of showing you a card just before you would forget it, to help you memorise almost anything: languages, exam material, medical terms, historical dates. The software runs on Windows, macOS and Linux, with a companion mobile app, and everything stays on your own device unless you choose to sync.

Bob: What makes Mnemosyne stand out is the research backing. The project collects anonymous performance data, with permission, to refine its scheduling algorithm, and users often report they can build a working vocabulary in a new language faster than with commercial apps. It handles images, audio and mathematical formulas, so it works for language learners, medical students, and anyone preparing for a certification exam. There are no ads, no subscription, and no account required.

Samantha: Setup takes about ten minutes. You download the program, create your first deck of cards, and start reviewing. The app then decides when to show each card again based on how well you recalled it. Users often describe it as feeling almost too easy at first, until they realise weeks later that the material has genuinely stuck. For anyone taking language classes, preparing for a European driving theory test, or studying for professional qualifications, it is a genuinely powerful tool that costs nothing.

Bob: One practical note for European users. Because everything stays local by default, there is no question about where your data goes or which cloud provider stores it. If you want to sync between devices, you can set up your own server, which is unusual but genuinely private. Reviewers consistently praise its longevity, some users have decks going back fifteen years.

Corporate AI Is Leaving the Lab — Now Companies Have to Prove It Actually Works

Samantha: Alright, let's talk about something that doesn't get enough airtime in AI coverage, which is the messy, unglamorous question of — okay, we've run all these AI pilots, we've spent a lot of money, now what do we actually have to show for it?

Bob: This is the conversation happening at Reuters Momentum AI in Austin this week, and I think it's genuinely one of the more honest public discussions the business world has had about AI in a while. The phrase being thrown around is the "prove-it phase" — the moment where experimentation has to give way to actual measurable outcomes.

Samantha: And one of the more interesting ideas came from KPMG's Swami Chandrasekaran, who introduced this metric called "cost per accepted output." Which basically asks — what does it actually cost you to get one usable result from your AI system, including all the human checking, error correction, and oversight on top of it?

Bob: That's a much more honest number than most companies are using. A lot of AI ROI calculations conveniently ignore the human labour that sits around the model making sure it doesn't produce something embarrassing or wrong.

Samantha: Exactly. And then you had people from FedEx and Mars essentially saying — if a project isn't delivering value, stop it. Which sounds obvious, but in practice companies have been very reluctant to pull the plug on AI initiatives because there's still this political pressure internally to be seen as doing AI.

Bob: The sunk cost problem, applied to machine learning. There's also the point about organisational structure — several speakers flagged that the reason AI projects fail isn't always the technology, it's that the company isn't set up to actually absorb and act on what the AI is producing.

Samantha: You can have the best model in the world, but if the workflow around it is broken, you're just automating a mess.

Bob: Pretty much. And I think what's encouraging about this conference moment is that we're hearing senior executives talk publicly about failure and about discipline — not just hype. That's a more mature conversation.

Samantha: It's the difference between AI as a story you tell investors and AI as something that actually changes how your business runs. And clearly a lot of companies are still bridging that gap.

Bob: Want to know more or react? Visit stateoftech.eu or email us at info@doorzetters.net. State of Tech, the tech world in 15 minutes.