TL;DR: On November 21, 2025, the FBI posted a Request for Information on SAM.gov seeking AI and machine learning systems for its drone fleet. The wish list: real-time facial recognition, license plate readers, weapon detection, and perimeter analysis, all running on NVIDIA Jetson Orin edge processors so identification happens onboard, no cloud connection needed. This comes as police drone flights have exploded (NYPD alone flew 6,546 missions in the first half of 2025, a 3,200% increase since 2022) and DHS has already deployed Predator drones over domestic protests. No warrant is required.
What the FBI Is Shopping For
The filing appeared on SAM.gov (the federal procurement portal) on November 21, 2025. The FBI's Operational Technology Division wants vendors to pitch AI and machine learning systems for unmanned aerial systems.[1]
The spec sheet reads like a surveillance wish list:
- Facial recognition: Identify individuals from live drone camera feeds
- License plate recognition: Read plates from the air in real time
- Weapon detection: Spot firearms via AI object detection
- Vehicle, vessel, and animal detection: Track anything that moves
- Perimeter analysis with directional awareness: Monitor who enters and exits a defined area
The technical requirements are specific. The FBI wants systems built on YOLO (You Only Look Once) object-detection frameworks, capable of decoding KLV metadata and Cursor-on-Target data from live camera feeds. Models must reach Technology Readiness Level 7, meaning a working prototype tested in real-world conditions, not a lab experiment.[2]
Most importantly: systems must run on NVIDIA Jetson Orin edge processors, deployed directly on the drone. That means facial recognition happens in the air, on the aircraft, in real time. No cloud uploads. No latency. No data trail to subpoena.[1]
Why Onboard Processing Changes Everything
Most surveillance cameras send footage to a server for analysis. That creates records. Logs. Subpoena-able data. Accountability, however imperfect.
Edge processing flips that model. When the AI runs on the drone itself, identification happens in real time with no external data trail. The drone sees your face, matches it against a watchlist, and flags you, all before anyone on the ground knows what happened.
For law enforcement, this is a feature. For civil liberties, it's a nightmare. If there's no server log showing who was scanned, how do you challenge a false match? How do you prove the FBI was running facial recognition at a protest if the processing happened on a chip that was flying 400 feet above your head?
Police Drones Are Already Everywhere
The FBI's RFI didn't land in a vacuum. Police drone usage has exploded.
New York City police conducted 6,546 drone flights in just the first six months of 2025. That's a 3,200% increase since 2022.[3] The NYPD used drones to monitor "No Kings" protests in October 2025, with minimal oversight to verify constitutional compliance.[1]
The FAA approved 410 drone-as-first-responder (DFR) waivers in the first two months after updating its process in May 2025, nearly a third of all DFR waivers ever granted.[4]
The companies building these systems aren't just selling cameras with propellers. They're building an integrated surveillance ecosystem:
- Flock Safety added automated license plate readers to drones[4]
- Axon and Skydio expanded their drone partnership into new police departments
- Brinc partnered with Motorola Solutions on DFR programs
- Paladin teamed with SkyeBrowse for 3D environmental mapping from drone feeds
These drone programs are increasingly feeding into real-time crime centers: the centralized surveillance hubs where police aggregate camera feeds, license plate data, social media monitoring, and ShotSpotter alerts. Adding AI-powered drone feeds means another data stream pouring into the same funnel.[4]
They've Already Used Drones Against Protesters
This isn't theoretical. The federal government has already pointed drones at Americans exercising their First Amendment rights.
In May 2020, as protests erupted in Minneapolis after the murder of George Floyd, the Department of Homeland Security deployed unmanned vehicles to record footage of demonstrators. DHS then expanded drone surveillance to at least 15 cities.[1]
In June 2025, DHS flew MQ-9 Predator drones (the same model used in overseas military operations) over anti-ICE protests in Los Angeles.[1]
In 2021, the U.S. Marshals Service used drones to surveil protesters in Washington, D.C.[1]
None of these deployments required warrants. None faced meaningful judicial oversight. The legal theory: airspace is public, so anything observable from the air is fair game. Now imagine those same drones running real-time facial recognition.
The Weapon Detection Problem
The FBI's RFI specifically requests AI-based firearm detection from drone feeds. Here's the problem: no company has proven this technology actually works.[1]
Matthew Guariglia, policy analyst at the Electronic Frontier Foundation, flagged the danger: a drone "whirling around the sky at an awkward angle" generating false weapon detections could trigger armed police responses to non-threats.[1]
Think about what that means in practice. An AI algorithm on a drone misidentifies a phone, umbrella, or tool as a firearm. Armed officers respond to a "weapon detected" alert. The person holding the object has no idea they've been flagged. In a country where police shoot first and investigate later, AI false positives from the sky aren't an abstract concern. They're a body count waiting to happen.
“Indiscriminate Mass Surveillance of All People”
Guariglia put it plainly: "By their very nature, these technologies are not built to spy on a specific person who is under criminal investigation. They are built to do indiscriminate mass surveillance of all people, leaving people that are politically involved and marginalized even more vulnerable to state harassment."[1]
That's the key distinction. A warrant targets one person based on probable cause. A facial recognition drone scans every face in its camera view. You can't run facial recognition on a crowd without scanning the faces of everyone in that crowd, including people with no criminal history, no warrant, no suspicion.
"One fear is police flying face recognition drones over protests to identify attendees," Guariglia added.[1]
That fear isn't paranoia. DHS already flew drones over protests. The FBI already has the nation's largest facial recognition database (with over 640 million face photos as of a 2023 GAO audit). This RFI connects those two dots.
Some Cities Are Fighting Back
Not everyone's rolling over.
Syracuse lawmakers blocked their police drone program four times in 2025, demanding privacy protections before deployment. Portland, Maine rejected a police drone purchase despite legitimate search-and-rescue justifications. The Sonoma County ACLU lawsuit could set legal boundaries for drone surveillance nationwide.[4]
But these are exceptions. Most drone programs launch with minimal public input, vague policies, and promises of restraint that evaporate the moment the hardware is operational. The United States has roughly 18,000 independent law enforcement agencies. Each one can set its own drone surveillance policies. There's no federal standard.
The FBI's RFI signals where the federal government wants this technology to go. Once the FBI deploys AI-powered facial recognition drones, local police departments will follow. That's how surveillance technology trickles down: the feds buy it first, then local agencies get grants to buy the same systems.
What to Watch
Follow the Contract
The RFI is the first step. A formal Request for Proposals (RFP) will follow. Track SAM.gov filings from the FBI's Operational Technology Division to see which vendors win the contract and what capabilities they deliver.
Fourth Amendment Challenges
Courts haven't ruled on whether AI-powered facial recognition from drones constitutes a "search" under the Fourth Amendment. The Supreme Court's 2018 Carpenter ruling said warrantless location tracking violates the Fourth Amendment. Does warrantless facial identification from the sky?
Local Drone Ordinances
Check whether your city or county has adopted drone surveillance restrictions. Organizations like the EFF's Atlas of Surveillance track police drone programs nationwide.
State Legislation
Several states are considering drone surveillance bills. Push for requirements that police obtain warrants before deploying facial recognition from aircraft, and mandatory public disclosure of drone flight logs.
The Bottom Line
The FBI wants drones that can identify your face from the sky, read your license plate, and detect whether you're carrying a weapon, all processed onboard so there's no server log to subpoena.
They've already flown drones over American protests. They already have the largest facial recognition database in the country. This RFI is the next step: merge the database with the drones and add real-time AI.
No warrant. No judicial oversight. No notification. Just a machine in the sky that knows who you are.
The technology companies invited to respond have until the deadline to pitch their capabilities. And somewhere in a procurement office, the FBI's Operational Technology Division is evaluating which system can scan the most faces per second from the most altitude for the least money.
Your sky isn't yours anymore. It might be time to look up.
References
- The Intercept: The FBI Wants AI Surveillance Drones With Facial Recognition (November 21, 2025)
- DronXL: FBI Seeks AI-Powered Surveillance Drones With Facial Recognition (November 23, 2025)
- S.T.O.P.: Surveillance Technology Oversight Project Response (December 2025)
- EFF: Drone First Responder Programs: 2025 in Review (December 2025)
- Biometric Update: FBI Issues RFI on Drones With Real-Time Facial Recognition (November 2025)