AI Could Turn ATF’s 4473 Stockpile Into the Gun Registry Congress Banned

American gun owners should be up in arms over any federal gun registration scheme. The threat of artificial intelligence only makes the danger worse.

For decades, Second Amendment supporters have warned that ATF’s growing stockpile of firearm transaction records could become the national gun registry Congress explicitly banned in 1986. ATF’s excuse has always been that its records are not searchable by a buyer’s name. That was never much comfort. In the age of AI, it is almost meaningless.

That is the warning at the center of a new Firearms Research Center paper by Del Schlangen, titled “Congress Banned a Gun Registry; AI Inference May Render the Prohibition Obsolete.” The paper argues that the long fight over ATF’s digitized firearm transaction records may already be behind the technology curve. The old question was whether ATF’s database is “searchable.” The new question is whether artificial intelligence can make that excuse irrelevant.

AmmoLand readers know the background. We have been covering ATF’s out-of-business dealer record stockpile for years, including the agency’s 920-million-plus record holdings, the digitization of those records, and the obvious concern that the Bureau is sitting on the building blocks of a national gun registry. AmmoLand previously reported that ATF acknowledged holding 920,664,765 firearm records, with almost all already scanned or digitized.

That number has only become harder to dismiss. In 2026, AmmoLand covered Senate testimony warning that ATF has nearly one billion firearm records, with 94 percent already digitized. GOA’s Erich Pratt told senators the system is “not a registry in name only” but “a confiscation list waiting to be used.”

ATF’s defense has been just as familiar: the records supposedly cannot be searched by purchaser name. That was never good enough. Now, in an AI world, it borders on absurd.

Congress banned federal gun registration in the Firearm Owners’ Protection Act of 1986. The law, codified at 18 U.S.C. § 926(a), prohibits “any system of registration of firearms, firearms owners, or firearms transactions.”

Schlangen’s paper focuses on “any system” because the Bureau’s argument has always leaned on technical hair-splitting. Former ATF Director Steven Dettelbach said ATF pays to remove search functionality from Adobe files to comply with Congress’s ban. Schlangen points out what that really means: ATF has the underlying data. Names, addresses, dates of birth, firearm descriptions, and serial numbers are visually present in the files. The claimed safeguard is a disabled software feature.

That is not a constitutional firewall. It is a software toggle.

The paper says ATF’s Out-of-Business Records Imaging System, or OBRIS, holds about 921 million digitized transaction records from dealers who closed or surrendered their licenses. These are images of Form 4473s, the forms Americans fill out when buying guns from licensed dealers. They contain the exact information gun owners have always feared the government would centralize: name, address, date of birth, firearm description, and serial number.

The old ATF talking point assumes that preventing text-based search prevents data extraction. Schlangen argues that modern AI destroys that assumption. Multimodal AI systems can process document images as visual objects. They do not need a normal PDF search bar. They can read, extract, structure, and connect data from images. A handwritten serial number on a 4473 is not invisible to AI just because Adobe search is disabled. That should end the “not searchable” excuse.

ATF does not need a neat alphabetical list to have a registry problem. If AI can extract names and serial numbers from scanned 4473s, connect them to other records, and produce person-to-firearm associations, then the government has created the functional equivalent of what Congress banned.

Schlangen calls this “registry-equivalent knowledge.”

The concept is simple. A traditional registry tells the government who owns what. An AI-inferred registry may do the same thing without being stored as a traditional registry. The paper identifies three dangerous capabilities: the ability to type in a person’s name and receive a likely firearms profile; the ability to generate lists of probable gun owners in a geographic area or population group; and the ability to connect a firearm, serial number, model, or ballistic clue to a likely owner outside the traditional trace process.

Any one of those should alarm every gun owner in America. All three together would be a shadow registry in everything but name.

It is a warning that the data, the technology, and federal AI policy are converging. Waiting until the switch is flipped would be a reckless way to defend the Second Amendment.

The danger also goes beyond OBRIS. ATF maintains multiple sales reports. It uses eTrace for firearm tracing. NIBIN contains millions of ballistic images. The NFA registry already tracks suppressors, machine guns, short-barreled rifles, and other restricted items. Each of those systems may be defended as limited, separate, or lawful on its own. But AI is built to connect dots.

A billion records, processed through AI, tied to trace data, ballistic data, dealer records, and other government inputs, can produce something far more dangerous than any single filing cabinet.

It can produce a gun-owner dossier.

This is where ATF’s long-running registry problem gets more serious. AmmoLand has already covered the Bureau’s push to digitize massive volumes of private firearm records and the concern that the agency is moving one step closer to a national gun registry. Prior coverage noted that Biden-era record-retention changes ended the old 20-year destruction limit and moved FFL records toward permanent retention.

AI makes that permanent stockpile more dangerous.

Schlangen also explains why the legal landscape is better for future challenges than it once was. The old NRA v. Reno case upheld temporary NICS record retention, but that involved a limited audit-retention system, not a permanent billion-record archive that could be processed with modern AI. More importantly, Reno was decided under Chevron deference, when courts often deferred to federal agencies. After Loper Bright, courts no longer have to accept ATF’s self-serving interpretation of what counts as “any system of registration.”

A post-Loper Bright court can read § 926(a) for itself. If Congress said “any system,” courts do not have to pretend the law only covers a database that looks like it came from 1986. A repeatable AI process that uses OBRIS images, eTrace data, multiple-sale reports, and other inputs to output firearm ownership associations is a system in the ordinary meaning of the word.

The Second Amendment concern is just as obvious. Schlangen points to First Amendment cases recognizing that government knowledge of constitutionally protected activity can chill the exercise of rights. The paper acknowledges that courts have not yet fully extended that chilling-effect doctrine to the Second Amendment, but the principle should not be hard to understand.

If Americans believe the federal government can quietly build firearm ownership profiles from dealer records, background-check residue, trace systems, and AI inference, some will hesitate before buying, selling, or keeping arms. That chill is not hypothetical. Registration has always been sold as paperwork. In practice, it tells the government who has what.

Once the government knows who has what, the next step is targeting for confiscation.

Gun owners do not oppose registries because they are paranoid. They oppose registries because history keeps proving them right. Registration is how governments move from regulating arms to tracking them. Tracking is how they move from tracking to confiscation, selective enforcement, or political intimidation.

The paper’s conclusion is blunt. ATF is a federal agency. Its firearm transaction records are government data. Federal AI policy is moving toward broad data ingestion. One of the publicly known barriers between ATF’s records and a searchable registry is disabled search functionality. In an age of OCR, entity extraction, record linkage, and AI inference, that is a fragile shield for a statutory command.

Congress needs to close the gap now.

The fix should be direct: amend § 926(a) so the registry ban applies not only to formal databases, but also to any AI system, machine-learning model, computational process, or inferential tool that allows the government to derive firearm ownership information from federal data. Congress or the Executive Branch should also wall off firearm transaction records from federal AI-ingestion and model-training programs. ATF should be forced to publish a public AI-governance policy and explain what it is doing with gun-owner data.

No registry means no registry.

Not a paper registry.

Not a searchable registry.

Not a billion-record “non-searchable” archive.

And not an AI-powered shadow registry built from 4473s, trace data, ballistic records, multiple-sale reports, or any other government stockpile of lawful gun-owner information.

ATF Gun Registry Exposed, Senate Hearing Raises Alarm Over 1 Billion Records

About Duncan Johnson:

Duncan Johnson is a lifelong firearms enthusiast and unwavering defender of the Second Amendment—where “shall not be infringed” means exactly what it says. A graduate of George Mason University, he enjoys competing in local USPSA and multi-gun competitions whenever he’s not covering the latest in gun rights and firearm policy. Duncan is a regular contributor to AmmoLand News and serves as part of the editorial team responsible for AmmoLand’s daily gun-rights reporting and industry coverage.