A Walking Tour of Surveillance Infrastructure In San Francisco

When you walk around the city, you'll notice almost everyone is looking down at their phones. I do it myself sometimes. While we're looking down at our own devices, we rarely notice the devices that are looking down at us.

If you look up, you start to notice just how often you're being watched. Inspired by a 2019 Seattle piece, I set out to map the different types of surveillance tech in San Francisco as of 2026. I wanted to know more about where these devices are and what they do. Who owns them? Where is the data stored and how can it be used?

Come with me on a walking tour of surveillance infrastructure in San Francisco. We'll pan, tilt, and zoom our way through the city, stopping by bullet cameras, turrets, face scanners, Flocks, StringRays, robotaxis, domes, and police drones. We’ll also stop at a few famous historic sites, including the buildings of both the watchers and the watchdogs. Our tour ends at the RTIC, the real-time aggregator used by SFPD's own surveillance network.

Every piece of tech in this network was approved individually as a sensible fix to a specific problem. Individually, many of them seem harmless. But together, they feed a networked system we never would have approved all at once. If at any point in our tour you feel uneasy, relax. We are doing this to protect you.

Our tour begins at the areas around Union Square and Yerba Buena Gardens. If you start around the Capital One Cafe and circle the block, you’ll notice just how many cameras there are. Security cameras owned by private businesses are of course nothing new, but these days the SFPD can pull in live feeds from non-city cameras. In September 2022, the San Francisco Board of Supervisors passed a controversial ordinance authorizing the SFPD to access privately-owned "non-city" surveillance cameras without a warrant, as long as the owner agrees.

Modern security cameras are also much more capable. Take, for instance, the Avigilon camera above. Its AI software classifies what it sees and flags "unusual activity." Maybe someone is standing around too long or walking around the block aimlessly (things I do all of the time). It may be useful to follow that person around. It also has facial recognition and “appearance search” functionality. While San Francisco has banned the use of facial recognition technology by the city itself, private owners are free to use the technology as they please.

In 2019, San Francisco passed the Surveillance Technology Ordinance (SF Admin Code Section 19B). The ordinance requires every piece of surveillance tech used by a city department to pass an impact report. Surveillance tech needs review by the Committee on Information Technology (COIT), the department head, the City Attorney, and a full Board of Supervisors vote. It also requires an annual public report on use of the technology. Below is a table listing of some of the technology in the inventory.

Notice that the surveillance inventory is still growing. I noticed the inventory has an "Upcoming" entry: a DataWorks Plus photo system with facial comparison and tattoo recognition. DataWorks is the company whose face-matching software wrongfully put a man in jail for shoplifting. It sells the software using the tagline "Ready to upgrade to Justice 2.0?"

The review requirements for police surveillance have loosened since the 2019 ordinance. In 2024, voters passed Proposition E, which lets SFPD acquire and use new surveillance technology for up to a year before submitting any usage policy to the Board.

The EFF (a watchdog group) urged a no vote, since the measure lets police deploy "virtually any new surveillance technology they wished for a full year without any oversight, accountability, transparency, or semblance of democratic control." In 2024, the policy became deploy first, approve later.

Flock Safety is a “police tech” startup which just raised money at an 8.4 billion dollar valuation. Its cameras are license plate readers (ALPRs) that photograph every passing vehicle and log details like plate number, make, model, color, and distinguishing features. They build a searchable location history for any car that can be accessed by California law enforcement agencies. San Francisco has 400 of them spread across roughly 100 intersections citywide, funded by a $17.3 million State Organized Retail Theft Grant.

On a walk through my neighborhood (Inner Richmond), I spotted 3 of them at the intersection of California Ave and Arguello, which I photographed and circled in red. Once you know what they look like, you start seeing them everywhere.

As I walked around town, I noticed them near many restaurants, cafes, and bars. Below are a few photos I took from my walks near Tony’s Cable Car, Lava Java, Madrone Art Bar, and Mucky Duck.

When I walked over to Alamo Square, it was a bit strange seeing these postcard views of San Francisco with cameras hovering overhead everywhere.

While the Flock cameras are focused on watching cars, there are many other types of cameras around Alamo Square, including several in front of the Painted Ladies.

I also began noticing all of the cameras in Golden Gate Park, including near places we just chill out. Places like the Shakespeare Garden and the Skatin’ Place roller rink where people skate and dance.

When I checked out the new Hardware Coffee in West Portal, I decided to pull up deflock.org, a crowd-sourced Flock map. To my surprise, there were several cameras nearby. I took a stroll through St. Francis Wood and spotted two almost immediately. DeFlock lists the operator as St Francis Wood HOA.

Flock markets to HOAs directly, so in this case, there was no vote or city involvement. The cameras were approved by the homeowners' board. From my research, it is already one of the safest neighborhoods in the city, with a median home price of around 4 million dollars. There is very little crime to watch in St. Francis Wood.

What’s funny is that during my walk, I was startled by a loud bang noise. It turned out it was a Lamborghini driving up the street. This street has a 25mph speed limit, but I doubt that this Lamborghini is going to get a ticket.

Just last week, I read about face scanners at Castro bars in The Gazetteer. I often visit Toad Hall with a couple of friends, but have always gotten right in without any checks at all. So I walked by on Thursday night to look again. I took one photo in front of Badlands across the street and checked out the website for this device. It is called Patronscan and is used to check IDs.

The device connects to a shared database of flagged patrons in participating venues. A person can be flagged at one venue and be automatically flagged across other locations, with instant alerts when a flagged person tries to enter. It is also able to track VIPs and “big spenders” so that staff can greet them and give special treatment when someone arrives. It also offers demographic data for marketing and a "value score" that rates the worth of the current crowd, using a method the company doesn't disclose.

One issue with facial recognitions systems is that they can be wrong. A Reno casino's system recently flagged an innocent man as a "100 percent match" for a banned patron. The police arrested him despite his valid IDs until a check proved the machine wrong.

Before Proposition E passed in 2024, San Francisco police had no drones at all. The program launched upon its passage and drone usage has only climbed since. It started with 8 flights in May 2024 and rose to 701 in a single month by February 2026. Under the Surveillance Technology Ordinance, SFPD posts every flight to DataSF, the city's open-data portal. The drone flight log includes date, duration, stated reason, and the block where it flew.

The data shows the fleet's biggest jump was in the fall of 2025, around the time the city announced the new real-time investigation center. Drone flights roughly tripled in a single month.

The map below uses the drone flight dataset and includes every logged flight since May 2024, fetched from data.sfgov.org. The drone flights cluster in SoMa, the Mission, and the Tenderloin.

The use cases for drones are still expanding as of this week. On June 9, 2026, the Board of Supervisors voted unanimously to let Public Works fly drones to catch people illegally dumping trash, and to identify who did it. The same vote created an "Illegal Dumping Camera System," which includes plate readers paired with pan-tilt-zoom cameras around the city. And just yesterday, Mission Local reported on a new drone-first-responder pilot in SoMa launching this Fall.

Enough about the tech for a moment. Let’s talk about some important sites in the history of surveillance. Below you’ll recognize City Hall, where the Board of Supervisors meet regularly. On the right is the Philip Burton Federal Building at 450 Golden Gate.

City Hall, of course, is where the facial recognition ban passed 8-1. They determined facial recognition technology endangers civil rights and civil liberties, and that these drawbacks outweigh its benefits:

Inside the Philip Burton Federal Building is the Northern California Regional Intelligence Center. The NCRIC is a fusion center connecting the SFPD to the FBI, DHS, DEA, and various Bay Area agencies.

In one documented 12-month window, NCRIC held 79.2 million license plate records from across the region, retained for a year, searchable by every member agency, including federal immigration enforcement. NCRIC also runs facial recognition, the capability San Francisco banned for itself. The ban, though, only applies to the city's own agencies. Since NCRIC is a regional center run jointly by federal and local agencies, the city’s ordinance doesn’t apply. So when SFPD wants a face identified, it can send the image to NCRIC, which runs the search and sends back a name. Because the software never runs on SFPD's machines, the department says it never "used" facial recognition. This workaround has been documented by the Washington Post and EFF.

Just a couple blocks north of the fusion center, at 815 Eddy Street, is the Electronic Frontier Foundation itself. A mile east, at 39 Drumm Street, is the ACLU of Northern California.

In 2018, the ACLU ran Amazon's face-recognition software against members of Congress and watched it falsely match 28 sitting members to mugshots, a disproportionate share of them people of color. The test made national news and helped make the case that the technology was too unreliable, and too biased to trust.

In 2025, the SF Standard exposed that the Flock cameras on this walk had been searched 1.6 million times by out-of-state agencies, including ICE, in violation of California law.

At 611 Folsom Street, AT&T runs a major internet switching center. In early 2003, AT&T installed fiber-optic splitters that copied internet traffic and fed it into a secure room, known as Room 641A. This room was accessible only to NSA-cleared staff.

In January 2006, Mark Klein (A former AT&T technician), brought documents showing this to the Electronic Frontier Foundation. His evidence led to the lawsuits Hepting v. AT&T and Jewel v. NSA. The courts dismissed the case on state-secrets grounds and never ruled on whether the surveillance was legal.

We end our tour at the control center of the operation: the eighth floor of 315 Montgomery. Until recently, this was the office of cryptocurrency company Ripple. Today, it houses the SFPD’s Real-Time Investigation Center (RTIC), which opened on December 3, 2025.

The RTIC was privately funded by Ripple’s co-founder, Chris Larsen, who subleased the space to the police through the end of 2026. With the SFPD running roughly 500 officers short, this room is pitched as a "force multiplier". The idea is to let hardware make up for the lack of officers.

Inside, officers monitor a live wall of data. Dashboards aggregate 911 dispatch calls, "hot list" hits from the city's Flock cameras, live video from the expanding drone fleet, and real-time feeds from thousands of cameras. If they need more, they can tap into a wider network of over 15,000 registered private cameras upon request.

While we often discuss the individual gadgets (eg. "Is a license plate reader really that invasive?"), that may be the wrong question. An individual license plate reader or a drone is just a single device. The RTIC brings them all together into a unified, aggregated system that is greater than the sum of its parts.

When I was growing up, you went out to play with your friends without being watched. We didn't have a network of cameras logging our every move, or AI flagging our "unusual activity." You just played outside, and you came home when the streetlights came on. Somehow we managed to live our lives just fine.

Our tour stops here, but the applications of surveillance tech are still growing. New technology is added to the surveillance inventory and new use cases emerge. None of these devices seem particularly alarming on their own. But when you map it all out, you can’t help but look at the whole thing and wonder: do we really need all this?