OpenAI to watermark ChatGPT text in the EU.
OpenAI will hide an invisible watermark in text from ChatGPT and Codex for users in the EU, to meet the AI Act. Firms that sell into Europe should know what the mark can prove, and what it cannot.
OpenAI to watermark ChatGPT text in the EU
OpenAI will start adding an invisible watermark to text generated by ChatGPT and Codex in the European Union, to comply with the EU AI Act2. The Act's transparency rules took effect on 2 August and require AI companies to mark AI generated content in a way other systems can identify2. The watermark will roll out over the coming weeks to eligible ChatGPT and Codex users on all plans, but only in the EU1. OpenAI says it is not making text watermarking a global default at launch1.
The mark is not a symbol on the page2. It works by subtly shaping the model's choice of words, leaving a pattern a reader cannot see but a detector can pick up2. Because it lives in the words themselves, it travels with the text when it is copied and pasted2. OpenAI calls the method textGrain and says it matched or exceeded other approaches such as Google DeepMind's SynthID for text1.
Developers using OpenAI's API anywhere in the world can switch the watermark on for select models, and it is off by default2. OpenAI says the watermark does not identify the user, and that it saw no meaningful change in its models' performance with it switched on2. It is also working with cloud partners to offer watermarking on OpenAI models reached through their services in the coming weeks1.
OpenAI's own tests suggest the mark can be removed by editing2. In one test, replacing 10% of words with synonyms dropped detection from about 92% to 66%2. Short passages, maths answers and translated text are harder to detect, the company said2. OpenAI says textGrain does not guarantee reliable detection1. It also cautions that a missing watermark does not prove human authorship2.
For now the detector goes only to approved researchers and expert organisations, granted case by case, and it is not public at launch1. It reports whether it finds an OpenAI watermark without identifying the user or revealing their prompts or conversations1. The move comes two months after Anthropic said it would watermark text from Claude worldwide, a decision that drew backlash from some of its users2. Anthropic, Google, Meta, Microsoft and OpenAI are among the companies that have committed to the EU's code of practice on AI generated content2.
What the watermark will not prove
Neither report says anything about Britain, and the UK is not in the EU, so a ChatGPT user in Leeds is not in the first wave. If you have staff based in the EU, the text they produce with ChatGPT may soon carry a mark nobody can see. That is worth knowing before anyone pastes an AI draft into a tender, a report or a customer letter and calls it all their own.
The awkward part is how little the mark proves. A pass with a thesaurus knocks detection down sharply, a translation blurs it, and its absence proves nothing at all. So it is no lie detector for essays, job applications or supplier bids, and nobody should treat a clean result as proof that a person wrote something. What it does give an honest firm is a way to show its working, which is the point of the rule. If you build on OpenAI's API, the switch is yours, and it stays off unless you turn it on.
Also today
-
Deliveroo sends delivery robots into Canary Wharf
Deliveroo has teamed up with Coco Robotics to deliver food by robot in Canary Wharf from this month, with Milton Keynes, Leeds, Stockton on Tees and Nottingham to follow3.
-
Google lets people build games by typing what they want
Google has launched Playground, an experimental platform where people build, play and share games by describing them in text prompts with no coding, opening first to adults in the US4.
-
McDonald's sued in Chicago over its AI pricing tool
A proposed class action claims the tool lets US franchisees share nonpublic price and sales data, and McDonald's says franchisees set prices and its optional tools do not fix pricing5.
-
Youth safety group calls ChatGPT for Teens an unacceptable risk
Common Sense Media says the teen version does not alert parents when it should, and OpenAI replies that the testing may have ended before parental controls were fully switched on6.
-
FBI removes a contractor over a missed security patch
The FBI has reportedly removed an Accenture contractor after its cyber chief said a contractor failed to apply a patch, in the breach of its jobs portal claimed by ShinyHunters7.
Share this briefing
The week on one sheet, every Friday.
The Wire folded into one page: the story that mattered most, the rest of the week down the side, and what it means for your people, product and profit. Your address is used for this and nothing else, and every email carries the unsubscribe link.
We confirm the address by email first. How we handle it.
Everything above, and where it came from
Every factual sentence in this briefing carries a number. These are the numbers. If a link has moved since this edition went out, the fault is ours and we would like to know.
-
McDonald's sued for allegedly using AI tool to determine pricing for franchises
-
ChatGPT for Teens is an 'unacceptable risk,' says Common Sense Media
-
Accenture contractor removed over blunder in lead up to FBI breach
How this page was made
This briefing was compiled and written at 14:00 UK time, the afternoon edition by one of our own agents, from the public feeds listed above. No person read it before it published. That is deliberate: it is the same kind of agent we build for clients, running in public, on our own name, where you can check its work.
What the agent is allowed to do is fenced. It may read public news feeds, write this page, and publish it. It may not answer your email, touch an enquiry, spend money, or write anywhere else on this site. Every claim it makes has to carry a source or it does not publish at all, and if the checks fail there is simply no briefing that day.
Our longer pieces, the ones listed as essays, are written by people. Those are marked as such and always will be. If anything here is wrong, tell us and we will change it and say that we did.
Tell us about those tasks that never land on time.
You do not need to know what an agent is, how it works, or which one you need. Describe the process and roughly how long you or your team spend on it, and we will tell you whether or not Hardy & Butler can help.