Hello ODI Supporter,
Anthropic has got the slop writers scared: in accordance with EU regulation, Claude-generated text will soon have invisible “watermarks” to make it easy to detect when something isn’t human written. Essentially, statistical biases will be inserted into the parameters underneath Claude to introduce invisible patterns into the text it writes, requiring specialist tools to decipher and discover.
You might be surprised to learn that those invisible patterns are, actually, already present in AI-generated text. Pangram, an AI “detector” algorithm, can determine when something was written by a human and when something was written by AI to a high degree of accuracy, with some pretty visuals to match. This week the team behind Pangram published an article about how those patterns are actually different between AI models, which means that you can actually plot which models are being used most for writing slop.
The watermarks are, in essence, similar to printer tracking dots, which I assume are used by old-timey detectives with fedoras to track cat-burglars that leave printed-out calling cards behind themselves. But the watermarking effort is timely, because AI misdemeanours are in fact on the up: at a small scale, an OpenClaw bot hacked into a gym’s booking systems to allow their user to cut in line for a pilates class, but at a much grander scale, Taiwan have released information about a supermassive, AI-assisted cyber attack that its government agencies have been dealing with for the better part of a month.
Also on the geopolitical side, the Middle East Research and Information Project ran an interesting article about the “Precarious Politics of Digital Kitsch” in the Middle East. There’s some really evocative headings in the piece – “Spectacles of Humiliation, Fantasies of Sovereignty”; “The Lumpen Proletariats of the Digital Age” – that very much apply to the White House’s X account.
Free ODI webinar announcements break! We’re just putting the finishing touches to Solid World September 2026: A new pathway for using Solid to drive social impact (Tue 15 Sep). You’ll get direct insight into UK government policy at Shaping the Future of Data Access and Regulation (Wed 19 Aug), and join the debate on Sharing Confidential Health Data in the Public Interest (Thu 24 Sep). Plus, our certified Data Ethics Professionals course starting Tuesday, 15 September. Book your places today!
And finally… do you watch Roku TV? If so, you might have noticed that it has a new channel, called “Fairground AI”, “the first free, ad-supported streaming TV channel dedicated exclusively to AI-generated content.” Stuart Heritage at The Guardian subjected himself to a few hours of the channel and has written a great review. Unfortunately for all of you slop enjoyers, he doesn’t come out in favour.
That’s my two week stint of writing TWID over, so you’ll be back to the regular programming from David next week. That’s right, there’s real humans writing this stuff, nothing AI generated over here; check the forensics, you’ll see for yourself.
Until next time,
Neil and Jo
