Every week, new models line up at the starting blocks, acquisitions reshuffle the deck, and European regulation tries, step by step, to keep pace. Meanwhile, you're left wondering where to actually find the signal in all that noise. That's exactly why we're launching a new series on the Xylos blog: What's new in GenAI land. Every two weeks, a sharp and honest overview of what's really moving in the world of generative AI, with the context you need.

This series is based on the biweekly LinkedIn overview by Tom Van 't veld, Learning Innovator at OASE (powered by Xylos). Tom has been closely following AI developments and sharing his findings for a while now. On the Xylos blog, we bring you a selection of the most relevant insights, focused on what matters for organizations and the people who work in them. Welcome to edition one.

No single player dominated the news over the past two weeks as much as Anthropic, and not always in a positive sense. But Google, Microsoft and European regulators also made their presence felt. Below are the six stories that really stand out.

Claude Cowork left preview and is now generally available for enterprises. Tom's tests were impressive, especially considering this is still just the end-user layer. OpenAI countered quickly with a new ChatGPT Pro subscription at 103 euros a month, aimed specifically at programmers. The battle for the professional market is in full swing.

At the same time, Anthropic announced Claude Mythos, and that announcement is… peculiar. The company had an outside psychiatrist conduct twenty hours of psychodynamic sessions with the model, because Anthropic doesn't rule out that very powerful AIs might have something resembling an inner experience. Whether that was really necessary or more a successful way to make headlines is hard to say. The rest of the message does sound familiar: just as with OpenAI's GPT-2 in 2019, the model will not be released publicly for now because it would supposedly be 'too dangerous'. We've seen this movie script before.

Then there was also the source code. Anthropic accidentally made the entire source code and architecture of Claude Code public, suddenly giving competitors a detailed look inside one of their core products. Through thousands of copyright takedowns, the company tried to get the code removed from GitHub, but developers are rewriting it in other programming languages and republishing it. The leak also revealed a spicy detail: Claude Code explicitly logs signs of user frustration, such as swearing and negative reactions, while at the same time scrubbing references to itself from the generated code so it comes across as more human.

The announcement of Claude Mythos also caused a stir beyond the tech world. The Atlantic argues that Mythos isn't just an internal Anthropic matter but a societal risk: the model would reportedly be able to independently discover thousands of serious vulnerabilities in virtually all common operating systems and browsers, a capability that until recently was reserved for state-sponsored hackers. The US Treasury Secretary and the Federal Reserve convened top executives from the largest banks for emergency talks, fearing a new systemic risk to the financial sector. Gary Marcus offers some nuance: the announcement was presented too theatrically, and the actual risks are less acute than suggested.

Gemini is getting notebooks that let you bundle chats, files and instructions per project and sync them with NotebookLM. At the same time, the Gemini app can now generate interactive 3D models and adjustable charts, letting you tweak variables or rotate objects directly in the chat. What was still vibe coding during a talk Tom gave a while back is now simply built in and accessible without any technical background. That's still a bit mindblowing.

The third update is an AI inbox in Gmail that prioritizes and summarizes emails, but for now it's only available with the Google AI Ultra subscription at 274.99 euros a month. On that front, Microsoft Copilot still has the edge.

Word is getting new Copilot features for legal and compliance professionals: word-level Track Changes, contextual comments and real-time progress views. Promising additions, although some features, like the Frontier preview, are still very much in rollout and not yet available everywhere.

That fits the pace at which Microsoft is broadly integrating AI across its product range: the Copilot name now appears in around 80 products and features, showing just how much is moving at once. For users and IT managers, that requires some orientation to keep track, and that's exactly where guidance can make the difference.

The European Commission is investigating whether ChatGPT should be classified as a 'very large online search engine' under the Digital Services Act. With more than 45 million monthly EU users, OpenAI is comfortably above the threshold, which, if the finding is positive, would mean mandatory risk assessments, audits and potentially hefty fines.

Meanwhile, Microsoft announces that from April 17, Copilot may process data outside the EU under heavy load, unless administrators actively disable this. Whether this is a genuinely new risk or is already covered by existing EU-US arrangements is currently disputed.

Research by Emerce shows that chatbots and voice assistants correctly answer only 12% of customer questions. Users also often struggle to reach a human employee. At the same time, 20% of full-time working Americans say, in a survey by Epoch AI and Ipsos, that AI has taken over tasks from them. The nuance the headline leaves out: it's almost exclusively about summarizing, rewriting, looking up information and preparatory analysis. Tasks, in short, that most employees weren't exactly losing sleep over anyway.

That brings Tom to a broader observation we'd like to close with. At the Dutch Enterprise Chamber, a lawyer once again ended up in the dock after she cited case law that simply doesn't exist. Regulators are logging more and more of these kinds of reports. Critical thinking and proofreading are basic skills that AI doesn't replace, but that the AI era may make more important than ever.

We'll be back in two weeks with the next edition.

About Tom Van 't veld

Tom has worked at Xylos for years, where he started as a Microsoft Office trainer and grew into the driving force behind innovative learning concepts. He laid the foundation for OASE, Xylos' online learning platform, and developed, among other things, the Digital Coach concept, a Microsoft Teams Escape Room app, and the mAindset game that lets employees learn to work with AI in a playful way. As Learning Innovator, he has increasingly focused in recent years on what AI means for the way we learn and work. Want to react or chat further? Find him on LinkedIn.