For 2 years they have stood in front of cameras explaining that AI is dangerous and must be controlled. On Tuesday they install one on every platform in the country.
The law is the TAKE IT DOWN Act. President Trump signed it on May 19, 2025. The compliance deadline is May 19, 2026. The House passed it 409 to 2. The Senate passed it on a voice vote.
The pitch is a 14 year old girl whose classmate generated nude images of her with a phone app. Her name is Elliston Berry. The bill is named for her case. The bill is also named for everything else the system it forces every platform in the country to install will be aimed at.
The law states that there must be “reasonable efforts” to identify and remove every known identical copy of flagged content within 48 hours, on a Federal Trade Commission fine of $53,088 per violation per day.
At the scale of Meta, Reddit, Snapchat, or X, that requirement is not satisfiable by human eyes. It is not satisfiable by deterministic database lookups. It is only satisfiable by perceptual hashing models and image classifiers trained on billions of images, matched against a database the platform does not maintain and the public is not allowed to see.
The mandate is the scanner. The scanner is AI. The law forces AI into every platform you use.
The AI in your hands is what the bill claims to fight. The AI in their hands is what the bill quietly requires. The model that generated the image of the girl is open source. It is on a 1,000+ mirrors. It cannot be retrieved. The bill does not try. It builds something else.
It builds a federal pipeline of automated scanners on every public-facing server in the country and points them at every photograph the citizen uploads, on a clock the platform is running scared of, audited by a body whose dissent has already been removed.
The infrastructure does not know what is in the database. The infrastructure does not care. PhotoDNA was sold to fight child abuse material in 2009. PhotoDNA now hashes terrorism content, copyright violations, leaked documents, and whatever sits in the operating list on a given Tuesday.
Once a perceptual-hash scanner is running on a platform, every new category of forbidden image is one database update away. The platform does not see the update. The user does not see the update. The list grows continuously.
The bill does not exempt encrypted services. The Electronic Frontier Foundation, the Center for Democracy and Technology, the Authors Guild, and a dozen other free-speech organizations asked Congress to write an encryption carve-out. Congress refused.
The remaining option is client-side scanning, which means the AI runs on your phone, on the photograph in your camera roll, before the message has a chance to be encrypted.
Apple tried this already in 2021. They withdrew it under public revolt. They are now being walked back to it under a different banner with a federal fine attached. The architecture is the same. The justification is always different.
Platforms are not required to disclose which AI scanners they run. The classifiers are not required to publish their false-positive rates. The users whose uploads are silently blocked are not required to be notified. The FTC, now a single-party body after the two Democratic commissioners were fired in the first weeks of the second Trump term, decides which platforms are sufficiently compliant.
By Tuesday morning every American server with an upload field has a federal AI inside it, trained by parties the citizen cannot identify, matched against a list the public cannot read, audited by a body whose dissent has been fired, on a 48 hour clock that runs against the platform and a fine that runs forever.
The first hashes are deepfaked teenagers. The next hashes are leaked documents. The hashes after that are whatever the chairman of the FTC would prefer not be online.
Move what matters off the platforms that will obey this monstrosity. Learn the tools that do not.
origamiaround
gamechangershow