AI facial recognition is the new doxxing accelerant. Masks do not help.

What's happening

Cops have always assumed a face mask, sunglasses, or a turned head was enough cover. That assumption is dead.

Between August and November 2025, ICE agents working enforcement details across the country got identified by name and home address from photos taken at protests and arrests. Politico ran the investigation in late August. Senator Blackburn sent formal letters to facial recognition vendors asking how it happened. The story kept growing through the fall.

The tool of choice is PimEyes. Subscription runs about $30 a month. You upload a photo of a face. The site returns links to every publicly indexed page on the internet that contains a matching face. News articles. Wedding announcements. LinkedIn. Department websites. A youth-soccer team photo from 2017. Any of those can carry a name. Once you have a name, a $20 broker lookup gives you a home address.

The mask does not help as much as cops think. PimEyes and similar tools work off facial geometry that is mostly visible above a surgical mask or a balaclava that leaves the eyes exposed. Sunglasses help more than a mask. A baseball cap pulled low helps more than either. None of those are sure things.

There is no federal law that prohibits commercial facial recognition for identifying a cop. A handful of states (Illinois, Texas, Washington) have biometric privacy statutes. None of them squarely prohibit a private subscriber from running a face match against public photos for harassment purposes. The FTC has opened inquiries. Nothing has produced a binding rule.

How the pipeline works in plain English

Walk through it the way it actually plays out.

1. Someone takes a photo of an officer at a protest, an arrest, a traffic stop. Phone cameras now resolve faces clearly at 30 feet.
2. The photographer uploads the face to PimEyes or Clearview's consumer-facing analog.
3. The tool returns links. Maybe a department photo from a 2019 promotion ceremony. Maybe a charity 5K. Maybe a wedding the officer attended where the bride's photographer posted galleries publicly.
4. One of those pages has the officer's name in a caption.
5. The doxxer takes the name to a broker site. Spokeo, Whitepages, BeenVerified, TruePeopleSearch. Twenty bucks gets a home address, sometimes a phone number, sometimes a list of relatives.
6. The doxxer posts the address on a Telegram channel, a 4chan thread, an "ICE List" wiki, or a private group chat.

That is the entire chain. Phone camera to home address in under an hour. No insider source. No hacking. No technical skill.

The ICE agent doxxing wave used exactly that pipeline. The Politico investigation traced it step by step.

Why it matters for cops, firefighters, and EMS

Three things.

First, the threat model changed. Photo OPSEC used to mean "do not post your patrol-car photo on Instagram." Now it means "any photo of you, anywhere, taken by anyone, can be the input." A mutual-aid call where a news photographer shoots the scene. A community-policing barbecue where a parent posts pictures. A retirement ceremony on the department website. All of those become breadcrumbs.

Second, the family threat angle is sharper than ever. Once a doxxer has the officer's name, they can run the same tool on family members. Wife at her gym. Kid at a school assembly. Husband at his employer's "meet the team" page. Any face in the immediate household becomes a separate input that confirms the home address.

Third, broker opt-out is no longer the whole answer. It still matters. Your name has to lead to the address for the chain to close. If the broker pages are scrubbed, the chain breaks. But the front of the chain, the photo-to-name step, runs on a different infrastructure that broker opt-outs do not touch.

Where this stands as of April 2026

A few markers from the past nine months.

• Politico, August 29, 2025. Investigation documenting the use of PimEyes and related tools to identify ICE agents from protest photos. The piece named several agents who had been doxxed and traced the tools used.
• Senator Blackburn, November 2025. Sent formal correspondence to PimEyes, Clearview, and adjacent vendors asking how their tools were used in the ICE-doxxing wave and what controls existed. The vendors mostly answered that their tools are not designed for that use, that they restrict certain queries, and that abuse is prohibited by terms of service. None of those answers prevent abuse in practice.
• House and Senate hearings, December 2025 through February 2026. Multiple subcommittees took testimony on commercial facial recognition. No legislation has emerged.
• State-level activity. California, New York, and Massachusetts have proposals in committee. The strongest version would require facial recognition vendors to obtain consent from the subject before returning a match. None has passed.

There is no live federal restriction on running a cop's face through PimEyes. The vendor's terms of service prohibit it. Enforcement is voluntary.

What an officer should do

Department-level steps. These need supervisor and chief buy-in. Push for them.

1. Audit the department website. Old promotion photos with names in captions are the highest-value input for a doxxer. Most departments still have these. Most do not need to.
2. Review press-release practice. A graduation list with photos and full names is a one-stop shop. If the department wants to publish achievement, do it without the photo or with a group shot that does not isolate faces.
3. Body-worn camera and dashcam release policy. State public-records laws often force release of footage. Some states allow redaction of officer faces. Use those redactions where the law permits.

Personal-level steps. These are within an officer's control today.

1. Search yourself on PimEyes. The free tier returns blurred preview hits. Buy one month of the paid tier to see what is actually out there. Go through every link. Some of it you can take down (your own social media). Some of it you cannot (a news photo).
2. Family OPSEC. Talk to your spouse about what they post, what their employer posts, what their friends post. Wedding photographer galleries are a frequent leak. Ask the photographer to take the gallery down or password-protect it.
3. Kids' school directories and team photos. Most of these were posted with parental consent that nobody read carefully. Pull the consent.
4. Broker opt-out as the second line. Once the photo-to-name link is harder, breaking the name-to-address link finishes the protection. We file across broker sites and re-check biweekly. See our broker opt-out coverage for what's enforceable where.

The mask is not the answer. The answer is making sure the face does not lead to the name, and the name does not lead to the address. Both halves matter.

What to watch

• Federal facial-recognition legislation. The next twelve months will see at least one serious bill. Watch whether it covers commercial use specifically, not just government use.
• Vendor consent decrees. The FTC inquiry into PimEyes could produce a settlement that restricts how the tool can be queried.
• State biometric privacy expansion. Illinois's BIPA has been the model. New York's bill is the most likely next state.
• A high-profile case. The next mass-doxxing event of cops, judges, or federal agents will probably move legislation faster than any of the above. That is bleak but accurate.

For context on the broker side of this chain, see our broker opt-out coverage and the threat model behind people-search sites. The face-to-name front of the pipeline is newer territory and harder to scrub. The name-to-address back of the pipeline is what we work on every day.