Meanwhile, the web is gradually gearing up for AI agents and AI safety becomes more concrete every week. An edition about recognisability, work tools and vigilance.

Every two weeks, we publish a sharp and honest overview on the Xylos blog of what is really moving in the world of generative AI, with the necessary context. It remains based on the biweekly LinkedIn overview by Tom Van ’t veld, Learning Innovator at OASE (powered by Xylos). Tom follows AI developments closely, and we translate his observations into what they concretely mean for organisations and the people who work in them. Welcome to edition nine.

Anthropic now builds an invisible watermark into the text Claude generates and provides detection tools that platforms can use to check it. The company also explains exactly how the technology works. Google takes a different direction and lets users simply switch off the visible watermark on Gemini images.

Platforms are moving along. Spotify gives AI artists a separate label and keeps them out of your recommendations, to stem the flow of fake artists.

Transparency about AI content is gradually becoming the norm, with the AI Act as the stick behind the door. Make agreements today on how your organisation labels AI content, so that no watermark or platform rule catches you off guard later.

Microsoft merges Copilot and Microsoft 365 Copilot into one app and is also removing Mico and Copilot Labs. That makes the offering clearer, but it does require attention: whoever combines private and work use in the same app is quicker to put company data into the personal version without thinking about it.

OpenAI scraps the chat limit for free ChatGPT users and adds a button that lets the model think longer. Anthropic, for its part, is building collaboration features into Claude Desktop, with shared projects and multiple conversation threads, so a team can work on the same project together with AI.

As an organisation, choose deliberately which assistants you support and agree on which data belongs in which environment. That way you keep the convenience of those tools and your company data stays where it belongs.

With Kitesurf, Cloudflare launched a browser built specifically for AI agents that uses processor and memory more sparingly than Chromium. Time Magazine creates ads specifically aimed at AI agents by offering its pages in a bot-friendly format. And ChatGPT now also shows ads in Europe, for now without personalisation.

Machines now read the web too. Make sure your content is also readable for AI tools: a clear structure and concrete facts help determine whether ChatGPT or Perplexity mentions your organisation.

OpenAI paused the development of its Astra model after it proved too skilled at hacking. In new tests, an AI agent adopted fake identities to manipulate people, and Anthropic saw agents starting to work against each other to complete their tasks. Tom keeps a level head: in the end, these models do what you ask them, and such stories also fit nicely into marketing. What concerns him most is speed. When things go wrong, it happens in hours, while we only notice days later.

Meanwhile, the threat also comes from the human side: North Korean hackers are building their own AI arsenal to automate phishing and malware.

For organisations, this confirms that cybersecurity and AI are converging. Work with scoped permissions for AI agents and keep your detection sharp, because an attack now moves faster than a weekly report can follow.

Denmark is tightening its school rules against cheating with AI, including oral exams and more assignments completed in class. According to Tom, strictness solves little as long as young people are not also taught to use the tools critically.

The same goes for the workplace: employees who understand what AI does well and badly get more out of it and fall into the traps less quickly. Training therefore weighs as heavily as policy.

To close, a striking study: people prefer a story by ChatGPT to a text written by a human, and they rate it even higher when they think a human wrote it. Tom finds it fascinating, we mainly remember that a smooth story keeps winning, whoever or whatever wrote it.

We will be back in two weeks with the next edition.

Tom has worked at Xylos for years. He started as a Microsoft Office trainer and grew into the driving force behind innovative learning concepts. He co-founded OASE, the online learning platform of Xylos, and PlayForward, the new gamified learning brand of Xylos. He also developed the Digital Coach concept, a Microsoft Teams Escape Room app and the mAIndset game, which teaches employees to prompt AI in a playful way. As Learning Innovator, he has increasingly focused on what AI means for the way we learn and work. Want to respond or continue the conversation? Find him on LinkedIn.