Authenticity under pressure: AI slop and weak provenance

'Ask HN: Add flag for AI-generated articles' (48886741) hit a hard problem. Detectors are unreliable, voters often can't tell, and HN's new guideline bans generated comments but doesn't clearly cover submissions. One commenter said stigma on Lobsters made them feel unwelcome even though they don't generate articles. Cliff Stoll's 'Rumors of my death are slightly exaggerated' (48037336) showed the nastier edge: an AI-generated obituary, and a commenter asking how we know it isn't an AI simulation of Cliff.

The C2PA post 'How to hack time' (49946707) shows the proposed fix is leaky. C2PA allows arbitrary 'exclusions' of byte ranges from signature calculations, and commenters compared that to SAML's mistakes, noting the EU AI Act already forces providers to embed these manifests. Together the threads say the same thing: the web needs trust signals, and neither detection nor signing delivers one yet.

A related thread (the Rodin Museum 3D scan verdict, 49946355) covers the same ground from another angle, over who gets to control digital copies of real things.