Anthropic IPO filing + EU Digital Fairness Act

Samantha: Welcome to The State of Tech, The European Edition, Tuesday November ten, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: Anthropic files for a stock market debut and ships faster, cheaper Claude models. Brussels presents the Digital Fairness Act against addictive apps and chatbots. NVIDIA's chief executive faces an FBI inquiry. AI agents break into websites for about fifty dollars a go. And two things you can use today: the browser extension Unhook, and Mozilla's free HTTP Observatory. Let's start with Anthropic.

Anthropic files for a stock market debut and releases faster, cheaper Claude models.

Samantha: Anthropic has officially filed for an initial public offering. According to reports, the AI lab is seeking a valuation of between one trillion and one and a half trillion dollars. That would make it one of the largest tech debuts. The company also has new products to show. It released new versions of two of its Claude models, called Fable and Mythos. Anthropic says both are thirty percent faster and twenty percent cheaper than the versions before them, and that they went through extensive safety testing before release.

Bob: The two models have very different jobs. Fable is built for complex technical work, the kind of tasks developers and engineers hand to an AI assistant. Mythos is aimed at advanced cybersecurity work, and it will not be widely available. Anthropic is deploying it only with a select group of organisations. So one model goes out broadly, while the other stays behind a gate. Anthropic describes this as balancing new capabilities with careful development while competition between AI labs is intense.

Samantha: The timing is notable. A company heading for public markets has to show two things at once: that it can keep shipping products, and that it can keep costs down. A price cut of twenty percent speaks directly to the second point. Many businesses pay for AI based on how much they use it, so a cut that size can change which projects are worth running. The speed gain adds up as well. It means shorter waits when an assistant works through a long technical task.

Bob: It is worth separating what is confirmed from what is reported. The filing is official. The valuation range is what reports say Anthropic is seeking, and it is not a final price. That number can move a lot before any shares change hands. Investors will still read the response closely. A valuation in the trillion-dollar range is a very large bet on future growth, and the reaction from public markets will signal how much appetite there is for AI companies more broadly.

Samantha: For Europe, there are two angles. European investors are watching this filing as a bellwether for the global AI market. European companies that build on Claude stand to benefit most directly from lower prices and faster responses. There is also a link to a story later in this episode. Mythos is meant for cybersecurity defence, and we will hear how attackers are already using AI agents to break into websites cheaply. The same kind of capability is showing up on both sides.

Brussels presents the Digital Fairness Act to curb addictive apps and AI chatbots.

Bob: The European Union has presented its Digital Fairness Act, and it targets the ways apps keep you hooked. Under the plans, tech companies would have to give users controls to switch off features designed for compulsive engagement. Social media platforms are one target. The act also names AI systems directly, including chatbots that imitate personal relationships to keep people talking. Companies that fail to comply could face fines of up to six percent of their global revenue.

Samantha: Six percent of global revenue is a serious amount. For the largest platforms, it is a sum their boards cannot ignore. The act is designed to complement the existing Digital Services Act. According to the reporting, the new act focuses on the basic architecture of how engagement works, meaning the design choices that keep you scrolling a little longer or chatting for another round. The rules look at how the product is built, beyond what appears on the screen.

Bob: The chatbot provisions stand out. The act explicitly covers AI chatbots that imitate personal relationships, so AI companions fall under the same rules as social media feeds. For people in Europe, the practical promise is more control, through a switch that turns off features built to keep you engaged. The reporting does not give a date for when those obligations would apply. For now, this is a plan on the table and not yet a setting on your phone.

Samantha: For companies, the message is that redesigns may be coming. Any platform or AI provider operating in Europe would need to review features built for compulsive engagement and give users a way to turn them off. That is a product question as much as a legal one. It also connects to our lead story, because AI firms courting global investors still have to build products that fit Europe's rules. Later in the episode, there is a tool that lets you reduce one of the most engaging feeds today.

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.

NVIDIA chief executive Jensen Huang faces an FBI inquiry and a board investigation.

Bob: NVIDIA chief executive Jensen Huang is reportedly under investigation by the FBI and by his own company's board. It follows his admission of an extramarital affair with a woman whom media reports identify as a suspected Chinese intelligence asset. US President Donald Trump confirmed the FBI inquiry and said Huang denied knowing about her alleged espionage activities. NVIDIA's board has announced an independent inquiry. The board also says it has no reason to believe national security or company secrets were compromised.

Samantha: It helps to be precise about what is established. According to the reports, Huang has admitted the affair. The espionage element rests on media reports that describe her as a suspected asset. According to Trump, Huang denies knowing anything about that. None of this amounts to a finding of wrongdoing. Still, an FBI inquiry combined with a board investigation was enough to unsettle investors. NVIDIA's shares fell as the story spread, and the reporting describes significant turbulence for the world's most valuable chipmaker.

Bob: NVIDIA's position explains why this reaches beyond one company. It is the world's most valuable chipmaker, and its chips are critical for AI development around the world. A national security inquiry involving its chief executive, with a suspected Chinese intelligence link, puts geopolitics directly on the board's agenda. So far, the board says it has seen no sign that secrets were compromised. The open questions are whether the FBI reaches the same conclusion, and how long the uncertainty lasts.

Samantha: For Europe, the concern is practical. European companies and research labs rely on NVIDIA's chips for their AI work. Turbulence at the top of the company could affect the availability and pricing of those chips, and that would spread to many tech sectors here. This is a possible risk and has not happened yet. The things to watch are the outcome of the board's independent inquiry, any further statements from the FBI, and how investors react when trading resumes.

AI agents breached more than a hundred websites at about fifty dollars per hack.

Bob: Norwegian cybersecurity firm NorForge has uncovered a hacker who used AI agents to break into websites largely on autopilot. According to NorForge, the agents probed and breached more than one hundred websites, at an estimated cost of about fifty dollars per successful hack. These were not scripted attacks repeating the same steps. The agents kept exploring, adapting and improving their methods as they went.

Samantha: One example from NorForge's findings shows how this works. The agents first tried a direct attack on a site and failed. They then examined the infrastructure connected to that site, found weaknesses in the control panels used to manage it, and got in that way. A control panel is the admin dashboard used to configure and run a site, so a weakness there can give an attacker the same access as the front door key.

Bob: That behaviour marks the shift. A pre-scripted attack follows a fixed set of steps, and if the trick fails, it moves on. In this case, the software acted more like a patient human attacker and looked for another way in when the first route failed. It also did this cheaply. A cost of fifty dollars per breach puts this kind of attack within reach of people who previously lacked the skills or the budget.

Samantha: The bottom line is that attacks are becoming more sophisticated and more frequent at the same time. That is how the source describes the trend: AI is lowering the barrier to entry and quickly increasing both the complexity and the number of cyber threats. Agents that adapt on their own, at this price, can be aimed at small targets as easily as large ones. That makes this relevant to every website owner, including organisations without a security team.

Bob: For Europe, the warning is direct. European businesses and critical infrastructure face the same kind of autonomous attacks, and a European firm brought this case to light. The source points to an urgent need for AI-powered defences and stronger cybersecurity rules. It also connects back to the lead story, because Anthropic's restricted Mythos model is aimed at advanced cybersecurity work. Attackers and defenders are now turning to the same kind of tools.

The free extension Unhook hides YouTube recommendations, autoplay and other attention traps.

Samantha: To close out, two things you can actually use today, and the first connects to the Digital Fairness Act. Brussels wants switches that turn off attention-grabbing features. On YouTube, a free browser extension called Unhook offers some of that now. Once it is installed, you can hide the recommendations on the home page, the suggested videos in the sidebar, the end screens that point you to the next clip, and autoplay. You choose which parts stay and which disappear.

Bob: With Unhook, YouTube works more like a search box than a feed. You go in for the video you wanted, and the page no longer lines up ten more. The extension is known for its simple menu. You click the icon and get a list of switches, so you can turn off the home feed but keep comments, or hide Shorts and leave everything else as it is. The goal is to remove the cues that pull you into the next video.

Samantha: Setting it up takes about two minutes. Search for Unhook in your browser's extension store, add it, and click its icon to open the settings. It is free and available for Chrome and Firefox. There are a few limitations. It only works in a desktop browser, so the YouTube app on your phone stays the same. And as with any add-on, it is worth checking which permissions the extension requests before you install it.

Bob: Timing is what links this to today's news. The Digital Fairness Act is still a plan, and the reporting gives no date for when platforms would have to offer these controls. Unhook is available now. It only covers YouTube, so your other apps stay the same. If one particular feed tends to take up your evenings, this is a direct way to see what changes when the recommendations are gone.

Mozilla's free HTTP Observatory grades your website's basic security settings in seconds.

Samantha: The second tool follows from the NorForge story. If AI agents can search websites for weak spots at fifty dollars a go, it helps to know how your own site looks from the outside. Mozilla's HTTP Observatory is a free online scanner that checks one part of that picture. You type in your website's address, it runs a scan, and within moments you get a grade from A-plus down to F.

Bob: The grade is based on security headers, which are instructions a website sends to your browser about how to behave safely. One example tells the browser to load the site only over an encrypted connection. Another stops other sites from embedding the page. Missing headers make some attacks easier. The Observatory lists which headers are present and which are missing, explains what each one does, and gives tips for fixing gaps. It runs in your browser and does not require an account.

Samantha: This is useful for anyone running a small business site, a club website or a personal blog. A scan takes about a minute. You can send the report to whoever manages your hosting and ask them to work through the list. It makes sense to scan sites you own or manage, because the check only reads public settings and your own site is the one where you can act on the results.

Bob: The limitation is important. A good grade does not mean your site is safe. The Observatory checks configuration and does not look for the kind of weaknesses NorForge describes, such as vulnerable control panels or outdated software. Treat it as a first check, not a full security audit. Strong admin logins and up-to-date site software are separate tasks. Still, AI is making it cheaper for attackers to find missing basic settings, which gives owners a reason to find them first.

Samantha: Today we covered: Anthropic's IPO filing and its faster, cheaper Claude models, Brussels' Digital Fairness Act, the FBI inquiry into NVIDIA's Jensen Huang, AI agents breaking into websites cheaply, and two things to try yourself: Unhook and Mozilla's HTTP Observatory.

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.