Since August 2nd, a European text has required machine-generated content to be labelled. A large American lab is applying it worldwide.

On the very day it published the manual for its mark, a code repository appeared to erase it. Four thousand marks of interest in two days.

Except nothing here is demonstrable. Not that the removal works, not that the mark is there. The tool that would settle it was never released.

And the mark does not say what everybody assumes it says. A text you wrote yourself can come back marked.

Meanwhile a fully generated news channel goes live, keeps its label on screen, and flags its own mistakes. Nobody was forcing it to.

Plus a small impression struck into English silver since the fourteenth century, worth nothing without the room where you take it to be read.

Three signals. One thread. One question…

🎙️ Every week, SynthAIcast listens to the world's best AI podcasts so you don't have to. Synthesised in ten minutes, verified by a human, in English and French

👉 Also available in French: SynthIAcast

📩 synthiacast@icloud.com

🔗 SOURCES, BY SIGNAL

Signal 1 - the mark, and what it actually proves

• Anthropic - official documentation of the marking, 11 August

https://tinyurl.com/5bbffp53

• European Commission - official list of signatories, 31 July

https://tinyurl.com/fpyup586

━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Signal 2 - the eraser, and what nobody can prove

• implicator.ai - the repository, its timestamp, its counter, 13 August

https://tinyurl.com/yey247mf

• BleepingComputer - the removal ecosystem and the missing detector, 13 August

https://tinyurl.com/49d34852

━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Signal 3 - the generated channel that shows its label

• Variety - launch, cleared avatars, on-air error flagged, 12 August

https://tinyurl.com/4tn5m7z8

━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Perspective - mandatory on one side, voluntary on the other

• European Commission - signing the code is voluntary, marking is not

https://tinyurl.com/34rarn42

━━━━━━━━━━━━━━━━━━━━━━━━━━━━

🎧 GOING FURTHER

• Tech Café, 10 August - can AI be creative, with Bertrand Misonne (in French)

https://tinyurl.com/svvxv5s5

• TWIML AI, 12 August - why image generation takes more than bigger models

https://tinyurl.com/mr6ft6j8

Also cited in the episode, unlinked for space: Euronews, Forbes and The Next Web on the worldwide scope of the marking, Digital Trends on the removal repository.

🌐 https://synthiacast.com

Hosted on Ausha. See ausha.co/privacy-policy for more information.