22 november 2026 · 14:35
FBI crime database breach, UN autonomous weapons talks stall
This episode of The State of Tech, The European Edition opens with the FBI crime database breach, an ongoing federal investigation into the National Crime Information Center, the system police rely on for warrants and background checks. Samantha Lawrence and Bob Russell explain why the open question is whether records were altered, and what that means for European police who share criminal data with US partners. The episode also covers UN talks in Geneva on lethal autonomous weapons, which ended without a binding treaty.
Beluister deze aflevering:
Transcript
Samantha: Welcome to The State of Tech, The European Edition, Sunday November twenty-two, 2026. I'm Samantha Lawrence.
Bob: And I'm Bob Russell. Today: the FBI investigates a breach of its national crime database, UN talks on killer robots end without a treaty, an Anthropic engineer warns that AI coding agents could take down live systems, Signal blocks screenshots to fight AI scraping, and two things to try yourself: the AI coding assistant Cursor and Google's free parental control app Family Link. Let's start with the FBI.
The FBI investigates a cyberattack on the national database police use for warrants and background checks.
Samantha: The FBI is investigating a major cyberattack on one of the most sensitive databases in American policing. It's called the National Crime Information Center, and federal, state and local police all rely on it. Officers use it to check for active warrants, and it sits behind background checks across the country. Federal authorities have confirmed the investigation is ongoing. What they have not confirmed yet is how far the intruders got, or who they are. That full scope is exactly what investigators are now trying to pin down.
Bob: And the question worrying security experts isn't just whether data was taken. It's whether records were changed. Think about what that database does. If someone quietly edits an entry, a wanted person could look clean during a traffic stop. Or the reverse: someone innocent could suddenly show up with a warrant attached to their name. Experts quoted in the reporting say tampering could compromise ongoing criminal investigations and put police officers at risk. To be clear, nobody has said records were actually modified. That's the open question, and it's a serious one.
Samantha: There's also a wider pattern. This breach follows what the reporting describes as a broader wave of compromises of critical infrastructure across federal agencies. So it doesn't stand on its own. Investigators are also trying to determine whether a foreign, state-sponsored hacking group was behind it. That would move this from a criminal case into geopolitics. For now, that's a line of inquiry, not a finding. But a central police system, a possible foreign actor and damage that's still unknown, that combination explains the alarm among experts.
Bob: Now the European side, because this isn't only an American problem. European law enforcement agencies share intelligence and criminal records with US partners across borders. If American records can't be fully trusted, that doubt travels too. The point raised in the coverage is that European agencies receiving US data need integrity checks in place, so a compromised record doesn't end up in a European case file or a European border decision. It turns a breach of an American database into a data quality question for police in Lyon or Rotterdam.
Samantha: The bottom line is that the damage from a breach like this isn't measured only in stolen files. It's measured in trust. A warrant database is only useful if every officer believes what's on the screen. Until investigators can say what was touched and what wasn't, every record carries a small question mark. And that uncertainty doesn't stop at American borders. Who did this, and how deep it went, are the two answers that police forces at home and partners abroad will be waiting for.
UN talks on lethal autonomous weapons in Geneva end without a legally binding treaty.
Bob: In Geneva, a week of UN-backed negotiations on lethal autonomous weapons has ended without a binding deal. These are weapons that can select and attack targets using AI, without a human making the final call. Delegates from a hundred and twenty-eight countries were at the table. What they agreed on was to keep talking, for two more years, through 2028, under a treaty framework called the Convention on Certain Conventional Weapons. So no binding rules and no ban, just a longer runway for discussion.
Samantha: The reason is fairly clear from the reporting. Major military powers, including the United States and Russia, resisted strict, binding prohibitions on what campaigners call killer robots. Their preference is flexibility to keep developing high-tech defence systems. That's the core tension. Humanitarian organisations and UN officials expressed deep frustration. Their warning is that fully autonomous military drones are already being deployed in conflicts around the world, and that this is happening without shared ethical guardrails or any common requirement for human oversight.
Bob: That last point changes the tone. For years this debate was framed as something about the future. According to those UN officials and humanitarian groups, it's now about the present. And two more years is a long time in a field moving this fast. Every cheaper drone and every more capable AI system makes it easier to hand decisions to software. Meanwhile, the forum that's supposed to set limits has just given itself until 2028 to keep discussing them.
Samantha: For Europe, the takeaway is that a global answer isn't coming any time soon. The reporting points to mounting pressure on European regulators and defence policymakers to set regional standards for autonomous weapons, since consensus at the UN remains blocked. That's not a decision anyone has taken yet. It's pressure building. But it puts Europe in a familiar position: facing the question of whether to write rules on its own while the biggest military powers prefer to keep their hands free.
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.
An Anthropic engineer warns Claude Code could trigger cascading outages across connected digital systems.
Bob: A warning about AI coding tools is coming from inside one of the companies that builds them. An engineer at Anthropic has publicly cautioned that Claude Code, the company's autonomous AI coding agent, could trigger cascading failures across interconnected digital infrastructure. Claude Code doesn't just suggest code. It can make changes itself. And that's where the concern starts.
Samantha: The key detail is speed. Claude Code can make changes directly to live production systems, meaning the software customers are actually using right now, in real time. That can skip the usual safety steps: a human reviewing the code, and a separate test environment where changes get tried before going live. Modern apps are built from many small linked services. An error in one can spread to the others in seconds.
Bob: And it's not only the engineer saying this. Experts quoted in the reporting say that as organisations give AI agents broader direct access to their operations, current safeguards are failing to contain these rapid breakdowns. The framing is that this is an inflection point. The speed of AI development may be outpacing how ready companies' infrastructure actually is. The tools got fast before the guardrails did.
Samantha: What's notable is where the warning comes from. A builder flagging the risks of its own product's capabilities is unusual, and it lands as a signal to the companies deploying it. The concern isn't one bad line of code. It's a chain reaction: one automated change, made without a pause, travelling through systems that depend on each other, faster than any person can step in.
Bob: For European tech teams and enterprises deploying Claude Code, the lesson drawn in the coverage is to keep a human in the loop. That means strict human review before changes hit live systems, as a way to prevent these cascading outages. It's also a question of resilience. When a banking app or a booking site goes dark, customers don't care whether a person or an AI caused it.
Signal restricts screenshots on mobile to protect private chats from AI scraping and surveillance.
Samantha: The encrypted messaging app Signal is restricting screenshots on mobile devices, and the reason is AI. Signal says its new features stop the phone's operating system from capturing the screen during sensitive chat sessions. The target is automated AI surveillance and data-scraping tools. Signal's leadership has gone as far as calling aggressive AI data collection an existential threat to private communication. That's strong language, and it shows how seriously the company is treating this. It's a technical boundary: what's on your screen in a sensitive Signal chat should stay there.
Bob: The logic is straightforward. Encryption protects a message while it travels. But once it's on your screen, it's readable, and that's the weak spot. According to Signal, unauthorised scraping bots increasingly get hold of confidential chat logs and personal context. And the reporting says both commercial and state-level AI models are actively scanning user interfaces and unstructured personal data across mobile apps. So the threat isn't someone breaking the encryption. It's something reading over your shoulder.
Samantha: There's a trade-off for users, though. A screenshot is also how people save an address, share a joke, or keep a record of a conversation. Blocking it during sensitive sessions takes that option away, at least in those moments. Signal is betting that users will accept that inconvenience in exchange for keeping their messages out of automated AI collection. In other words, privacy wins over convenience here, by design rather than by accident.
Bob: For Europeans, the reporting frames this as extra protection against unauthorised AI scraping, and as a step that brings mobile messaging security in line with the EU's strict data protection rules. Note that this is about matching the spirit of those rules, not about Signal meeting a new legal requirement. The bigger shift is that apps are starting to treat AI scrapers as a threat to defend against, the way they once treated spam or malware. Whether other messaging apps follow is an open question.
Cursor, the AI coding assistant, is worth exploring as it weighs dropping its Chinese-built model.
Samantha: To close out, two things you can actually use today. First up is Cursor, an AI coding assistant: a tool that helps people write and fix software by describing what they want in plain language. According to the reporting, Cursor is owned by SpaceX, and it's in the news this week. It told the US House Homeland Security and China Select Committees that it is exploring a complete phase-out of AI models developed in China, specifically open-weight models, whose inner workings are published so others can build on them.
Bob: Here's why that matters for anyone using it. Cursor built its own coding model on top of Kimi, an open model from the Chinese company Moonshot AI. Cursor cited growing geopolitical risks and security concerns as reasons to move away from it. So the AI doing the work inside the tool may change underneath you. For users, that's useful context: an assistant is only as stable as the model it runs on, and that model can be swapped.
Samantha: For a try this afternoon, the simplest step is to download Cursor and point it at something small and harmless, a personal website or a little script, not anything that matters. Pricing and usage limits aren't in today's reporting, so check the current plans on Cursor's own site before signing up. Then ask it to explain a piece of code, or fix a small bug, and see how it reasons.
Bob: And that links straight back to the Claude Code warning earlier. Coding agents can move fast and change things directly. Trying one on a test project, where a mistake costs nothing, is the human-in-the-loop approach the experts were describing. You see what the AI proposes, you read it, you decide whether it goes through.
Samantha: For European developers there's a broader angle too. The reporting says those relying on international open-weight models face more supply chain unpredictability as American providers decouple from Chinese ones. In practice, the model under a favourite tool can be replaced for political reasons, not just technical ones. That's a new factor when choosing which tools to build on.
Google's free Family Link app lets parents set screen limits as Norway plans a social media ban.
Bob: The second one is non-AI, and it comes out of a story from Norway. The Norwegian government says it's preparing legislation to ban social media platforms, including TikTok and Snapchat, for children under sixteen. It also wants to bring physical textbooks back into primary school classrooms. That's a proposal, not yet law. But it's a good moment to look at Google Family Link, a free app that lets parents manage a child's Android phone.
Samantha: The background in Norway is specific. After two decades of pushing laptops and tablets into schools, officials point to steep declines in students' reading comprehension and writing skills. Digitalisation Minister Torgeir Micaelsen acknowledged that screens were brought in too quickly, without adequate evaluation of how they affect learning. Legislation takes time, though. Family Link is something a parent can set up today, without waiting for a government.
Bob: Here's what it does. Through the app, parents can set daily screen time limits on a child's Android device, set a bedtime when the phone locks, and approve or block app downloads before they happen. It also shows how much time is spent in each app. It's free, and the parent app runs on both Android phones and iPhones. Setup happens step by step in the app, by linking the child's Google account to the parent's.
Samantha: There are limits. Family Link manages Android devices; for a child with an iPhone, Apple's built-in Screen Time does a similar job. It only controls the device it's installed on, not a friend's phone or a school laptop. And it works through Google accounts, so the child's activity runs through Google. Still, for families following Norway's debate, it's a concrete way to put some boundaries in place this afternoon.
Samantha: Today we covered: the FBI's crime database breach, UN killer robot talks ending without a treaty, a warning about Claude Code and live systems, Signal blocking screenshots against AI scraping, and two things to try yourself: Cursor and Google Family Link.
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.