Topic
Techniques—statistical classifiers, style analysis, and watermarking—used to detect AI-generated text or other content, often unreliable.
The Claude watermark survives light editing and dissolves under heavy editing, and its absence proves nothing. Policies that treat it as a verdict are built on a binary that does not exist.
Perspective Coverage
6 publishersReality
Anthropic's Article 50 compliance mark is applied during decoding, so text Claude only edited is flagged too. There is no opt-out, no visibility, and no way to audit coverage.
Perspective Coverage
4 publishersReality
Anthropic says the mark exists to satisfy the EU AI Act and that it has no durable way to scope it by region, so every customer gets it. OpenAI's support page says text watermarks are coming too.
Perspective Coverage
3 publishersReality
Techdirt argues Anthropic complied with the EU AI Act more broadly than the law required. For anyone shipping model-written text, the vendor's implementation is the constraint that actually binds.
Reality
Graduate-degree viewers in a 315-person study identified the human-edited video 34.8% of the time, below a coin flip. The result rests on one pair of videos where the human editor added the polish, so what carries over is the advice to test content on real users.
Reality
The company showed 1,722 people a mix of human and AI-written comments in a simulated feed. The accounts they caught most reliably were the hostile ones that awareness training already teaches staff to distrust.
Reality
A salesperson said his manager asked him to point his LinkedIn page at the day job after the CEO wondered about split time. The career coach quoted in the same report says recruiters now find candidates by searching that page.
Reality
Gizmodo asked five AI systems whether a viral image of the 9/11 rescuer Welles Crowther was real. Each described the fabrication as a photograph of some kind, and what caught it was a reverse image search.
Reality
Two detection chief executives describe their own products as probabilities rather than verdicts. That is the constraint to design around before a score is allowed to fail a student or close an account.
Reality
The Hanover Institute has pushed out more than 100 unbylined, AI-written articles in under a month: question headlines, citations that do not link, and a crawler map for the models reading it.
Reality
LinkedIn's new report button collects human quality judgements at scale. Detectors and watermarks answer only who wrote a page, which leaves retrieval pipelines to score corroboration themselves.
Reality
Originality.ai flagged 1,272 of 2,034 books. The audit's soft spots and Amazon's honour-system disclosure point the same way: nobody is verifying who wrote what.
Reality
The standard signs a chain of assertions about capture and edits. It does not verify the scene in front of the lens, and a missing manifest tells you almost nothing.
Reality
Zvezdelina Stankova says she used AI only to edit her op-ed on Berkeley admissions. A Pangram reading of roughly 33% moved the story anyway, which is a byline problem, not a detection one.
Reality