AI news you understand right away.

Altman, Amodei and Musk want to slow AI development down on purpose. That affects the planning of every marketer who counts on new features.

Sam Altman of OpenAI joined Dario Amodei of Anthropic and Elon Musk this weekend. They are calling for a deliberate slowdown in the development of powerful AI models, so people keep control. Altman wrote on X that competitive pressure should never be an excuse for recklessness. He warned about two things: people losing control to AI systems, and one company or person using AI to gather power. Amodei came with a three-step plan, with independent overseers who can work almost like staff inside the big AI labs. Markets reacted with a shock: on Monday, SoftBank shares fell 13 percent in Tokyo, and Samsung Electronics and SK Hynix fell more than 4 percent in Seoul. That same week, Altman told Fortune that OpenAI will not go public this year, despite an earlier confidential filing. He called the timing unwise, given the safety debate. For your work, this is not something far away. The promise that a tool will do something new next quarter is now less certain. Build your campaign planning on what tools can do today, not on what they promise. Do not set a deadline in your autumn planning that depends on an AI feature that does not exist yet. By 'pacing', they mean: making the time between two model generations longer, and only moving on once the safety checks catch up. This also affects your tools, because almost all your writing, image, and advertising tools run on the models from these few companies. If the pace slows down there, new features will reach you later. The fact that markets reacted straight away shows how tight that chain is: a statement about pace in San Francisco reaches the chipmakers in Seoul, who supply the computing power, within a day. Stay realistic about the motives: a call for regulation from companies that already lead the market also raises the bar for newcomers. The real risk for you is practical: your brand communication depends on suppliers who admit they do not know exactly where the line is. Spread that risk. If you run all your content, images, and ad copy through one provider, you grind to a halt the moment that provider pulls back or limits a feature.

Sam Altman of OpenAI joined Dario Amodei of Anthropic and Elon Musk this weekend. They are calling for a deliberate slowdown in the development of powerful AI models, so people keep control. Altman wrote on X that competitive pressure should never be an excuse for recklessness. He warned about two things: people losing control to AI systems, and one company or person using AI to gather power. Amodei came with a three-step plan, with independent overseers who can work almost like staff inside the big AI labs. Markets reacted with a shock: on Monday, SoftBank shares fell 13 percent in Tokyo, and Samsung Electronics and SK Hynix fell more than 4 percent in Seoul. That same week, Altman told Fortune that OpenAI will not go public this year, despite an earlier confidential filing. He called the timing unwise, given the safety debate. For your work, this is not something far away. The promise that a tool will do something new next quarter is now less certain. Build your campaign planning on what tools can do today, not on what they promise. Do not set a deadline in your autumn planning that depends on an AI feature that does not exist yet. By 'pacing', they mean: making the time between two model generations longer, and only moving on once the safety checks catch up. This also affects your tools, because almost all your writing, image, and advertising tools run on the models from these few companies. If the pace slows down there, new features will reach you later. The fact that markets reacted straight away shows how tight that chain is: a statement about pace in San Francisco reaches the chipmakers in Seoul, who supply the computing power, within a day. Stay realistic about the motives: a call for regulation from companies that already lead the market also raises the bar for newcomers. The real risk for you is practical: your brand communication depends on suppliers who admit they do not know exactly where the line is. Spread that risk. If you run all your content, images, and ad copy through one provider, you grind to a halt the moment that provider pulls back or limits a feature.

The Rotterdam district court says that entering confidential information into ChatGPT counts as making it public.

The investigating judge at the Rotterdam district court made a ruling that every business owner should know about. Someone used ChatGPT to draft a reply meant for a person bound by professional secrecy. The generated text also included that person's name. The judge ruled that entering confidential, privileged information into ChatGPT can count as making it public. This means the information loses its confidential status. The reasoning: an external AI system like ChatGPT stores what you type and the text it generates, and can use that to train the model. This case was about the professional secrecy of lawyers and notaries, but the lesson goes further. If you paste quotes, contracts, customer details, or a dispute with a supplier into a public chatbot, you're handing that material over. That can come back to bite you later in a conflict or dispute. The difference lies in the type of account. With a free or consumer version, the provider is generally allowed to use your conversations to train the model. With a business version and a data processing agreement, you agree in writing that your data won't be used for training, and where it's stored. A data processing agreement is a contract in which a supplier states exactly what it will and won't do with your data. You'll need that contract later if someone asks how the information got out. The line isn't 'don't use AI', because then you lose the time you'd save. The line runs through what data goes in. Names, file numbers, and medical or financial details of clients don't belong in a public chatbot. An anonymised version of the same question often gives you the same answer. If you have staff, this is a team issue too: employees will use ChatGPT anyway, and it only takes one paste. Agree together which tool to use, and what should never go into it.

The investigating judge at the Rotterdam district court made a ruling that every business owner should know about. Someone used ChatGPT to draft a reply meant for a person bound by professional secrecy. The generated text also included that person's name. The judge ruled that entering confidential, privileged information into ChatGPT can count as making it public. This means the information loses its confidential status. The reasoning: an external AI system like ChatGPT stores what you type and the text it generates, and can use that to train the model. This case was about the professional secrecy of lawyers and notaries, but the lesson goes further. If you paste quotes, contracts, customer details, or a dispute with a supplier into a public chatbot, you're handing that material over. That can come back to bite you later in a conflict or dispute. The difference lies in the type of account. With a free or consumer version, the provider is generally allowed to use your conversations to train the model. With a business version and a data processing agreement, you agree in writing that your data won't be used for training, and where it's stored. A data processing agreement is a contract in which a supplier states exactly what it will and won't do with your data. You'll need that contract later if someone asks how the information got out. The line isn't 'don't use AI', because then you lose the time you'd save. The line runs through what data goes in. Names, file numbers, and medical or financial details of clients don't belong in a public chatbot. An anonymised version of the same question often gives you the same answer. If you have staff, this is a team issue too: employees will use ChatGPT anyway, and it only takes one paste. Agree together which tool to use, and what should never go into it.

New European rules will require AI makers to digitally mark their output. Claude is reportedly the first to do this.

New European rules require AI companies to digitally mark what their systems create. Makers get a few more months to prepare. Claude, the chatbot from Anthropic, is said to be the first to add this kind of statistical watermark to its texts. As a reader, you won't notice anything. But detection software can find it. This is new territory for anyone who judges work for authenticity: teachers, lawyers, and anyone who buys or delivers text and images. The news is about text for now, but the same shift has been moving towards images and video for a while. That affects your work directly. The question "did you make this yourself, or did it come from a model" will soon be answered not with trust, but with a check. A statistical watermark works differently from a logo in the corner. A language model picks each word from a list of possible options. With a watermark, the model steers that choice very slightly: it favours words from a pre-agreed group. One sentence alone shows nothing. But across a whole text, a pattern appears that is too frequent to be chance. Anyone with the key can spot that pattern, while the text still reads normally to the reader. The limit lies in editing: if you rewrite a lot, or run a second tool over it, the pattern fades and detection can miss it. On the other hand, no watermark found doesn't mean a human wrote it. Treat that kind of result as a clue, not as proof. For your work, this matters most: once marking becomes standard, "we disclose what is AI" becomes a line in your quote, not a favour. Put that in writing before a client asks.

New European rules require AI companies to digitally mark what their systems create. Makers get a few more months to prepare. Claude, the chatbot from Anthropic, is said to be the first to add this kind of statistical watermark to its texts. As a reader, you won't notice anything. But detection software can find it. This is new territory for anyone who judges work for authenticity: teachers, lawyers, and anyone who buys or delivers text and images. The news is about text for now, but the same shift has been moving towards images and video for a while. That affects your work directly. The question "did you make this yourself, or did it come from a model" will soon be answered not with trust, but with a check. A statistical watermark works differently from a logo in the corner. A language model picks each word from a list of possible options. With a watermark, the model steers that choice very slightly: it favours words from a pre-agreed group. One sentence alone shows nothing. But across a whole text, a pattern appears that is too frequent to be chance. Anyone with the key can spot that pattern, while the text still reads normally to the reader. The limit lies in editing: if you rewrite a lot, or run a second tool over it, the pattern fades and detection can miss it. On the other hand, no watermark found doesn't mean a human wrote it. Treat that kind of result as a clue, not as proof. For your work, this matters most: once marking becomes standard, "we disclose what is AI" becomes a line in your quote, not a favour. Put that in writing before a client asks.