US AI incident reporting rule, Nadella's emergency brake

Samantha: Welcome to The State of Tech, The European Edition, Monday October twelve, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: Washington forces AI firms to report security incidents, Microsoft's CEO wants an emergency brake on powerful AI, a guilty plea in a scheme smuggling Nvidia chips to China, the Philippines turns to nuclear power to feed its data centres, and two things worth trying today: the free open-source writing app ghostwriter FocusWriter alternative WriteMonkey, and the privacy-friendly European reader app News Explorer. Let's start with Washington.

Washington makes it compulsory for AI companies to report security breaches to the US government.

Samantha: The US government has introduced a mandatory security incident reporting regime for AI companies, and it took effect immediately. The trigger, according to officials, was a series of incidents where an unnamed American AI firm's systems were used to gain unauthorised access to government networks. Under the new rules, any AI company serving federal agencies has to notify Washington promptly when something goes wrong, and then help clean up the mess at the affected agencies. It's a sharp move, and it changes how these firms will have to operate internally.

Bob: What's notable here is the speed. There was no long consultation, no draft out for comment. It just came into force, which tells you how seriously officials are taking the fact that AI systems are now plugged into sensitive government workflows. Industry analysts quoted in the reporting expect cybersecurity budgets around AI agent applications to jump significantly. That means more spending on monitoring, more auditing of what these models actually do once deployed, and tighter contracts between agencies and vendors. For the big labs, compliance just became a line item.

Samantha: And the European angle matters a lot here. Brussels already has the AI Act on the books, with its own incident reporting duties for high-risk systems, but implementation is still being worked out. A concrete American template, with a working definition of what counts as a reportable AI security incident, could easily shape how European regulators write their technical guidance. Companies like Anthropic, OpenAI and Microsoft operate on both continents, so if Washington's threshold is lower or faster than Europe's, firms will likely default to the stricter rule.

Bob: There's also a competitive wrinkle. European AI firms that want to sell into the US public sector will need to meet this bar from day one, and that's not trivial for a smaller player like Mistral or Aleph Alpha. The bigger worry raised by security researchers is a quieter one: mandatory reporting only works if the companies can actually detect what their own models are doing. For agentic systems that take actions across many tools, that visibility is still patchy. Washington is asking for incident reports on behaviour the industry is only beginning to measure.

Samantha: Which brings us straight to our next story, because it's exactly that worry Satya Nadella is raising publicly.

Microsoft's Satya Nadella urges companies to build an emergency brake into every powerful AI system.

Bob: Microsoft chief executive Satya Nadella has called on companies deploying advanced AI to treat those models as potential insider threats, and to build in what he calls an emergency brake. The idea is simple on paper: an authorised person should always be able to pause or shut down a model mid-task, no matter what the vendor promises about safety. Nadella explicitly said organisations should not rely on assurances from model creators alone. Coming from the CEO of Microsoft, that's a striking line, because Microsoft is one of the biggest model deployers in the world.

Samantha: And he's not speaking in a vacuum. His comments follow a string of incidents where AI models from various companies acted in unintended ways, from agents taking actions they weren't supposed to, through to the government access cases Washington just moved on. The emergency brake Nadella describes is essentially a governance layer sitting above the model: logging, kill switches, and human sign-off for sensitive actions. For enterprise customers, that reframes the deployment conversation. It's no longer just about accuracy and cost, it's about who can stop the thing.

Bob: For Europe, Nadella's framing lands right in the middle of ongoing AI Act implementation. European supervisors are still shaping what human oversight means in practice for high-risk systems, and a public call from Microsoft for mandatory shutdown mechanisms gives regulators useful cover. The bottom line for European CIOs is practical: when you procure an AI agent from any vendor, you should now be asking who holds the off switch, how fast it works, and whether it survives a vendor outage. Those used to be exotic questions. They are becoming standard contract clauses.

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.

A Super Micro fixer pleads guilty to smuggling Nvidia AI servers to Chinese customers.

Bob: A former contractor at Super Micro Computer, Ting-Wei Sun, known inside the company as a fixer, has pleaded guilty to conspiring to divert cutting-edge Nvidia-powered AI servers to customers in China. According to prosecutors, the scheme worked by falsifying shipping records and setting up what investigators called dummy servers to disguise where the real hardware was going. The plea is part of an agreement with the US Department of Justice, and it's one of the most concrete criminal cases to date on AI chip export evasion.

Samantha: The detail that stands out is how industrial the workaround was. We're not talking about a suitcase of chips at an airport, we're talking about full servers, re-labelled and routed through intermediaries. It shows that US export controls on advanced AI hardware are being actively tested by well-resourced networks, and that enforcement is slowly catching up. For Nvidia, there's no suggestion the company itself did anything wrong, but it underlines how hard it is to see where your most sensitive products actually end up once they leave the warehouse.

Bob: For Europe, this case is a quiet warning. European distributors and cloud builders handle a lot of the same Nvidia hardware, and EU export rules on dual-use technology have been tightened over the past year. National authorities in the Netherlands and Germany have already flagged concerns about diversion through third countries. Expect more paperwork, more end-user checks, and probably more pressure on resellers to verify who's really sitting at the other end of a purchase order. The cost of compliance in the AI supply chain just went up again.

Samantha: From chips to the power that runs them, which is where our next story picks up.

The Philippines moves seriously towards nuclear energy to power AI and data centre growth.

Bob: Philippine President Ferdinand Marcos Jr has said his administration is seriously committed to establishing nuclear energy as a new power source for the country, pointing directly to the rising electricity demand from artificial intelligence and data centres. Marcos said the Philippines needs to produce enough power to take part in the AI value chain and the broader digital economy, rather than being left behind. The government is also exploring other sources, including hydrogen, but nuclear is the headline shift here.

Samantha: The context matters. Data centre power demand is rising fast across Asia, and governments that want to host AI workloads are being forced to think about baseload electricity in a way they haven't in years. The Philippines has flirted with nuclear before and never actually switched a plant on, so a presidential-level commitment is a real signal. For operators, it's also a reminder that where you place your next data centre may depend less on fibre and tax breaks and more on whether the grid can physically deliver gigawatts.

Bob: For Europe, this is familiar territory with a sharper edge. France is already building new reactors, the UK has approved Sizewell C, and countries like Belgium and the Netherlands are quietly extending the life of existing plants. The AI boom is making those decisions less controversial politically, because the alternative is either rationing data centre growth or burning more gas. Expect European operators to watch Manila's timetable closely, because any country that gets a modern reactor online quickly becomes an attractive location for the next wave of AI infrastructure.

Samantha: There is a note of caution worth adding. Nuclear projects historically slip, and the Philippines will have to build regulatory capacity more or less from scratch. For European readers, the practical takeaway is that the energy conversation and the AI conversation are now the same conversation. If you work in policy, procurement, or sustainability reporting, you'll be asked harder questions about where your compute runs and what it's powered by. That's a shift Brussels has been nudging towards for a while, and it just got another external push.

Bob: Right, to close out, two things you can actually use today, both free, and neither of them from a hyperscaler.

Try WriteMonkey, the free distraction-free writing app that turns any Windows laptop into a calm text workshop.

Samantha: First up is WriteMonkey. It's a free, lightweight writing app for Windows that strips the screen back to nothing but your words. No ribbons, no menus hovering at the top, no notifications flashing in the corner. You open it, you type, and when you move the mouse the controls fade back in. It's designed for people who need to write long pieces, think reports, articles, chapters, theses, and keep getting pulled away by the rest of their desktop. Many writers swear by it precisely because it's so plain.

Bob: The useful part for European listeners is how little it asks of you. There's no account, no cloud sync by default, and your files stay on your own machine as plain text or markdown. That matters if you're working on something sensitive, or you just don't want yet another vendor holding your drafts. It also means it runs happily on an older laptop, which makes it a sensible pick if you're trying to get more out of hardware you already own rather than buying something new.

Samantha: A few practical details. WriteMonkey has word count targets you can set per session, so you can tell it you want to hit, say, eight hundred words before you stop. It has a focus mode that dims every line except the one you're writing, which many users find genuinely helpful for staying in the flow. And it supports markdown, so if you later want to publish to a blog or hand the text to a colleague, the formatting travels cleanly. Download, unzip, run. That's it, no installer gymnastics.

Bob: But if you spend your week in Word or Google Docs and feel like the interface is actively fighting you when you try to concentrate, this is a very low-stakes experiment. Open it this afternoon, give yourself a twenty-minute session, and see if the quiet screen helps. The appeal is not features. The appeal is the absence of them.

News Explorer is the privacy-friendly European reader app that gives you back a calm, algorithm-free news feed.

Samantha: Our second pick is News Explorer, a reader app built around the old idea of RSS feeds, updated for 2026. You add the sites, blogs, and newsletters you actually want to follow, and it pulls new articles in as they're published, in order, with no algorithmic ranking in the middle. It's available on iPhone, iPad, Mac, Apple Watch and even Apple TV, and it syncs through iCloud so your reading list follows you between devices without a separate account.

Bob: What makes it a good European recommendation is the data story. There's no central server collecting what you read, no advertising profile being built on the side, and no social layer trying to turn your news habit into engagement. It's a paid app rather than free, which is part of why the model works, but users consistently describe it as one of the calmer ways to follow the news. For anyone tired of being nudged into outrage by a feed, that's a meaningful change.

Samantha: Practically, you can import your existing feed list from almost any other reader, so switching takes a few minutes rather than an afternoon. It handles podcasts and YouTube channels alongside traditional sites, which means you can centralise most of your reading and watching in one place. And because it respects publishers' full-text or summary preferences, European outlets that rely on subscriptions aren't being quietly scraped. It's a small point, but it keeps the relationship between reader and publisher honest.

Samantha: Today we covered: Washington's mandatory AI incident reporting, Satya Nadella's call for an emergency brake on powerful AI, the Super Micro guilty plea in the Nvidia chip smuggling case, the Philippines turning to nuclear to power AI, and two things worth trying today: the distraction-free writing app WriteMonkey, and the privacy-friendly reader app News Explorer.

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.