I know my usual beat is AltNPS, but since yesterday when this Politico article dropped, I've seen a lot of chatter about the ICE List, an activist project focused on unmasking ICE officers. That's a noble goal. ICE doesn't deserve to hide. But I want to encourage some healthy skepticism here.
It will be funny when Homeland kicks down their door and their (successful) defense is "we didn't actually expose anyone lmao".
Yeahhhh, I'm gonna have to see some data on that one lol.
First, there's a lot we don't know about this project. Project lead Dominick Skinner is short on details, and isn't even willing to name which LLM/AI model that his team is using to supposedly unmask ICE agents.
And given that by Skinner's own admission, the facial recognition tech has a SIXTY PERCENT ERROR RATE, I don't think this is something we should get behind.
This baffled me when I saw people discussing it like....we've been fighting the cops using this type of tech this whole time why do we magically think it works when used on them
The ICE List site says that "An AI research team," which goes unnamed, assists the organization, and that "the results speak for themselves." And yes, I suppose they do. The results are that it's wrong 60% of the time, and that's IF we take Skinner at his word.
![[From the ICE List website]: We also collaborate with several partner organizations, StopICE.net, ICE Spy, and other who would rather us not reveal the cooperation at this point. Their work and data are invaluable. An AI research team works with us, too. They’ve developed tools that identify agents who haven’t covered their faces or changed gear. They help spot patterns in behavior, movement, and appearance that humans might miss. Their work is technical, but the results speak for themselves. We verify agents using a layered system of confirmation: visual identification, behavioral matching, database records, and independent corroboration. Nothing makes it to the public without at least two solid confirmations. Three is preferred. It’s not just about accuracy. It’s about protection. For us, and for the truth.](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:7f5fdeb4ci3lefmt4zr3m3wg/bafkreig7sdqbuqk444uodqpsezyaia754srp3dacpja5vofwr4yiuydtoi@jpeg)
Skinner says that "a group of volunteers verifies [the AI-generated faces] through another process before posting any names online," but we don't know what that process is, who these volunteers are, or what their qualifications are. And the description on the ICE List's website isn't reassuring.
![[From the ICE List website]: Facial recognition helps, but it’s only one piece of the puzzle. Our volunteers are trained to spot recurring traits. We use side-by-side comparisons with verified past footage. We analyze background noise, street signs, timestamps, anything that helps establish context. Each confirmed ID goes through a second process. We document everything: date, time, location, appearance, clothing, gear, vehicle, any interaction with the public, and any verbal identification made. Then we check again. Double confirmation is our standard. Triple confirmation is our preference. We create individual agent profiles. These include all confirmed appearances, timestamps, and geographic movement. We track where agents have been deployed and what events they’ve been linked to. If a raid happened in Texas and that same agent appears at a courthouse in New York a month later, we note it. We work with volunteers who specialize in facial recognition. We maintain a private archive of footage. We use satellite imagery and street view tools to confirm raid locations. We log field offices, staging areas, known facilities, and temporary hubs. We document repeat appearances and trace patterns of movement. Our goal isn’t just to identify one agent. It’s to identify networks. Patterns. Roles.](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:7f5fdeb4ci3lefmt4zr3m3wg/bafkreidkezivbrbddoavuqj3fxsclwnax2o6l7vib5judfdjtqnx6ld26u@jpeg)
The "How We Identify Agents" section of the ICE List site claims that the org collects and analyzes a dizzying amount of data, but the public doesn't seem to have access to that data. Clicking a name on the list gives you an image (sometimes a stock image), and MAYBE a link to a LinkedIn profile.
The pages usually invite users to provide info about that agent, asking "What do you know about this agent?" Or "Have you seen this agent commit a crime?" But some pages don't even have that. This one is just a pic of a guy. No identifying info. No explanation of how they know this guy's name.
We don't know enough to trust this group. We don't know how they get their info. They don't let us see their data. If they have "trackers for license plates, uniform details, equipment loadouts, and agency cross-over," that feels like something an activist group would share with the public.
All we know is that they're using admittedly faulty AI tools to "unmask" ICE agents, and that they have not one but TWO convenient donation buttons on their homepage.
ICE doesn't deserve privacy, and the ICE List is a good goal in theory. But their caginess about their methods, their dramatic and grandiose tone, and their up-front requests for money make me skeptical. Please think twice before donating to this organization.
Thanks for this. Anyone who's done real research knows the best leads usually come from locals/people closest to targets. These lofty national projects raise big alarm bells and people need to be especially careful about IDing feds (as journalists, we have a *smidge* more legal protection there)
pretty typical politico it's just as sketchy to use for this as it is to let cops use it