Education edition
ChatGPT is getting a watermark that you cannot read yourself. A detector marked 48 of 50 theses as partly written by AI. GPT-6 can add buttons and other features to answers.
The watermark in ChatGPT text does not help with marking
OpenAI gives ChatGPT text in the EU an invisible watermark. Teachers cannot use the detector for now. The watermark is also easy to weaken.
OpenAI announced on Monday 5 October that text from ChatGPT and Codex in the EU will get an invisible watermark. Codex is OpenAI's tool for writing computer code. OpenAI is doing this to comply with Article 50 of the AI Act, the European law on AI. That article is about transparency and has applied since 2 August 2026. The technology is called textGrain. The watermark stays in place when you copy and paste text. The rollout will take the next few weeks and applies only in the EU. Eligible users will get the watermark with all subscriptions. Customers who use OpenAI through an API can switch on the watermark anywhere in the world. An API is a link that lets other software use the model. The watermark is off by default there. The next deadline is 2 December 2026. It applies to systems that were already on the market. For now, only approved researchers and organisations can use the detector. So you cannot read the watermark yourself. textGrain changes the model's choice of words very slightly. A language model chooses from several suitable words each time. If that choice is guided a little, it creates a pattern. You cannot see the pattern. A detector can recognise it. The pattern is in the words themselves. That is why it stays in place after copying and pasting. This is also its weak point. OpenAI says detection falls from almost 100% to below 20% if someone changes a quarter of the words. So a student can almost remove the watermark by rewriting a quarter of the words. A text without a watermark does not prove that a person wrote it. The watermark is only in text from ChatGPT and Codex, and only where it has been introduced. If you ever get access to the detector, a score is a clue, not proof. If you want to see what a student can do on their own, you could have part of the assignment written in class.
Background for subscribersA few reports touch your work directly. ChatGPT is getting a watermark, but teachers cannot use the detector themselves. A changed text can also lose the watermark. A detector also marked many UvA theses as AI text, but a score is not proof. GPT-6 can make answers with charts and tools look convincing, even when the content is wrong. A detector score is at most a reason to talk, not proof of cheating.
GPT-6 creates answers with charts, buttons and forms
OpenAI began rolling out GPT-6 in ChatGPT on 7 October 2026. Paying users with Plus, Pro, Business or Enterprise got GPT-6 Sol on 7 October. Users of the free version and Go followed on 8 October. They get the smaller GPT-6 Luna. GPT-6 includes 'Intelligent UI'. This lets ChatGPT build answers with charts, buttons, forms and diagrams. It can also include small tools, such as a calculator or a cost splitter. OpenAI says the model starts answering while it is still thinking. When it searches the web, the wait is 44% shorter than with GPT-5.6 Instant. OpenAI says it has more than 1.2 billion users a week. So the model does not just give you text. It can also create parts that you or your pupils can use. This could make an answer more useful, for example in an exercise with numbers. It could also make the answer seem more convincing. A polished diagram might look as if someone has already checked it. That does not mean the content is correct. Pupils may therefore see an answer that looks good more quickly, but is still wrong. If you want to see it for yourself, you could ask a question about your subject and check whether the chart or calculator is correct.
Background for subscribersThis is a taster.
Subscribers read the whole Education edition: every story, each with its background in plain language.
Subscribe →Not ready for that? Read the free letter first →
AI and tech, made simple.
This is the Education edition. Subscribers get the background to every story, in plain language.
Subscribe
inklaretaal · Amsterdam · © 2026