The Surveillance State: A Survival Guide

We live inside a machine that records. Your phone reports your location continuously. Cameras read your license plate as you drive to work. Your purchases, searches, messages, and movements are logged, sold, aggregated, and stored in buildings the size of shopping malls. Most of this happened in a single generation, and most of it happened with little public debate.

This article does two things. First, it traces how the modern surveillance apparatus was built, from the legal rupture after September 11th through the Snowden disclosures to the license plate readers now bolted to poles in thousands of American towns. Second, it lays out what an ordinary person can actually do to reduce their exposure. The goal is informed self-protection, the same project the Electronic Frontier Foundation, the ACLU, and privacy researchers have pursued for decades. None of this is about evading legitimate investigation. It is about the basic right to not be tracked by default.

Part One: How We Got Here

The Rupture: September 11th and the Legal Reset

Before 2001, domestic surveillance in the United States operated under meaningful constraints. The Foreign Intelligence Surveillance Act of 1978, passed in the wake of the Church Committee's exposure of decades of intelligence abuses, required the government to get warrants from a special court before wiretapping. The wall between foreign intelligence and domestic law enforcement was real.

The attacks of September 11th, 2001, dissolved that wall. Within six weeks Congress passed the USA PATRIOT Act, which dramatically expanded the government's authority to collect records, monitor communications, and share intelligence across agencies.

The 9/11 attacks triggered the legal reset that produced the modern surveillance apparatus

Section 215, the so-called "business records" provision, would later become the legal basis for the bulk collection of Americans' phone records. The Act passed the Senate 98 to 1.

What followed was a decade of secret programs built on aggressive and largely hidden interpretations of these new authorities. The public knew surveillance had expanded. It did not know how far.

The Architecture Revealed: Edward Snowden, 2013

In June 2013, a 29-year-old NSA contractor named Edward Snowden, working for Booz Allen Hamilton, handed a trove of classified documents to journalists Glenn Greenwald, Laura Poitras, and Barton Gellman. What those documents revealed reshaped the global conversation about privacy.

Edward Snowden speaks remotely at the Token2049 conference in Singapore on September 18, 2024. Suhaimi Abdullah/Bloomberg via Getty

The programs were extensive, and several names became permanent fixtures of the surveillance vocabulary.

PRISM allowed the NSA to collect data directly from the servers of the largest technology companies. Leaked materials named Microsoft (reportedly the first partner, in 2007), Google, Yahoo, Facebook, Apple, Microsoft's Skype, YouTube, and others. Under Section 702 of the FISA Amendments Act, analysts needed only a "reasonable belief," described in training materials as 51 percent confidence, that a target was a foreign national abroad.One of the leaked PRISM slides showing NSA collection partners, from the 2013 disclosures

Upstream collection tapped the physical fiber-optic cables that carry internet traffic, intercepting communications in transit rather than pulling them from company servers. An NSA training slide reportedly instructed analysts to use both PRISM and Upstream together.

XKeyscore was the search engine that sat on top of it all. The NSA's own training materials described it as the agency's "widest-reaching" system for developing intelligence from the internet. It indexed emails, online chats, browsing histories, and metadata, and let analysts query enormous databases by filling in a simple on-screen form with a broad justification, with no prior court authorization required for individual searches. Snowden's most famous claim, that he could "sitting at my desk" wiretap "anyone, from you or your accountant, to a federal judge or even the president," referred to capabilities like these. Officials denied the specific claim, but the XKeyscore documents lent it weight.

Boundless Informant was the NSA's tool for visualizing and counting its own collection. By March 2013, the agency had gathered 97 billion pieces of intelligence from computer networks worldwide in a single month.

The disclosures were not limited to the United States. The British agency GCHQ ran a program called Tempora that tapped fiber-optic cables and shared the take with the NSA, part of the "Five Eyes" intelligence-sharing arrangement between the US, UK, Canada, Australia, and New Zealand.

EFF aerial photograph of the NSA Utah Data Center in Bluffdale, the physical home of bulk collection. -Rick Bowmer / AP

The Warehouse: The Utah Data Center

All of that collected data needs to live somewhere. The physical answer sits in Bluffdale, Utah, on a National Guard base at the foot of the Wasatch Range. The NSA's Utah Data Center, code-named Bumblehive, spans roughly 1.5 million square feet, cost an estimated 1.5 to 1.7 billion dollars, and was completed around 2014. It draws about 65 megawatts of power continuously, enough to supply tens of thousands of homes, and contains four data halls plus extensive backup power and cooling.

The facility's exact storage capacity is classified. Estimates ranged wildly in 2013, from exabytes to the speculative "yottabyte" figure floated in early reporting. What matters is the principle: the center was deliberately designed with Moore's Law in mind, built to hold not just what the government collects now but what it will be able to collect as storage gets cheaper. The architecture assumes indefinite retention.

The Private Layer: Palantir

Government agencies do not build all their own tools. Increasingly, they buy them. The most significant private contractor in this space is Palantir Technologies, founded in 2003 with early funding from the CIA's venture arm, In-Q-Tel.

Palantir Technologies, the CIA-backed data analytics contractor whose government footprint has expanded sharply 

Palantir does not collect data in the way the NSA does. It builds the software that fuses data from many sources into a single searchable interface, letting an analyst pull together records that were previously scattered across incompatible systems. It has worked with US agencies since at least 2011. The ACLU has tracked the company since then, including its involvement in a predictive policing system in New Orleans that incorporated "threat scores."

The company's government footprint expanded sharply in recent years. In April 2025, ICE finalized a 30 million dollar contract with Palantir to build a system called ImmigrationOS, the Immigration Lifecycle Operating System, intended to provide "near real-time visibility" into the immigration lifecycle. According to reporting and congressional correspondence, the system draws on data from across the federal government, including the IRS, Social Security, and passport databases, alongside license plate reader data, to build profiles of enforcement targets. The broader sole-source contract it was folded into has reportedly grown past 145 million dollars, and Palantir has reportedly received roughly 900 million dollars in federal contracts since the start of the current administration. A related tool reportedly called ELITE has been described as helping map neighborhoods for enforcement operations.

The pattern is the principle that the ACLU has identified repeatedly: mass surveillance infrastructure, once built, expands beyond its original justification.

The Local Dragnet: Flock Safety and License Plate Readers

The most consequential recent development is not a secret government program. It is a private company selling cameras to your town council.

Flock Safety, founded in 2017 and headquartered in Atlanta, manufactures automated license plate readers (ALPRs), small solar-powered cameras typically mounted on poles near intersections and highway ramps. They photograph every passing vehicle, capturing the license plate plus what the company calls a "Vehicle Signature": make, color, body type, and distinguishing features like roof racks, bumper stickers, or body damage. Each read is tagged with time and location and uploaded to a centralized, searchable database.

A Flock Safety automated license plate reader on a utility pole

The scale is national. Flock provides ALPR technology to thousands of law enforcement agencies. In Oakland alone the company has installed nearly 300 cameras. The reads from local cameras flow into a shared network that agencies across the country can search.

The civil liberties concerns are concrete, not hypothetical. In 2025 the EFF analyzed 10 months of nationwide Flock searches and found that more than 50 federal, state, and local agencies ran hundreds of searches connected to protest activity, in some cases targeting known activist groups. The EFF also documented more than 80 agencies using search language that perpetuated harmful stereotypes against Romani people. In a widely reported case, a Texas police department searched Flock data in Washington state as part of a nationwide search for a woman who had an abortion. Flock data has been accessed by ICE. The company has announced it will let police request not just still photos but live video feeds and short clips.

Flock defends its system as constitutional, arguing that ALPRs create discrete, point-in-time observations at fixed locations rather than continuous tracking, and that the company requires users to log a reason for each search and maintains audit capability. Courts have so far largely accepted the point-in-time framing. But the trajectory, more cameras, more data types, longer retention, video on demand, and a single nationwide search interface, is exactly the mission creep critics warned about. Some states, including Washington, have begun introducing legislation to regulate ALPR data storage and access.

The Data Center Boom

Two forces are now converging. The surveillance and data-broker economy needs ever more storage and processing. The artificial intelligence boom needs the same thing, and the two markets increasingly use the same infrastructure. The current wave of hyperscale data center construction across the United States is driven primarily by AI compute demand, but the facilities that train language models are architecturally similar to the facilities that store and process surveillance data, and the line between commercial data aggregation, advertising surveillance, and government access has always been porous. The Snowden documents showed that commercial servers were a primary collection point. That has not changed.

The result is an environment in which the default state of an ordinary person's life is recorded, retained, and made queryable.

Part Two: How the Systems Work and What They Look For

You cannot protect yourself against a system you do not understand. Here is what the major surveillance modalities actually key on.

License Plate Readers

ALPRs use cameras plus optical character recognition (OCR) and machine learning. The OCR layer reads the plate characters. A second layer, the "vehicle signature" or "vehicle fingerprint," classifies the make, model, color, body style, and visible distinguishing marks so a vehicle can be identified even if the plate is obscured or misread. The system logs plate, signature, timestamp, and GPS location, then makes all of it searchable and cross-references it against "hotlists."

What it looks for: a specific plate, or a vehicle matching a description, appearing at a place and time, and patterns of movement assembled from multiple camera hits.

Facial Recognition

Facial recognition maps the geometry of a face into a numerical template, the distances and ratios between features, and matches that template against a database. Companies like Clearview AI built their databases by scraping billions of images from the public web and social media. ICE has contracted with Clearview and deployed field tools that scan faces and fingerprints.

What it looks for: a face that matches a template already in a database. The weak points are anything that disrupts the feature geometry the algorithm depends on, and, more practically, simply not having your image in the database in the first place.

Device and Network Tracking

Your phone is the most efficient tracking device most people carry. It broadcasts to cell towers continuously (cell-site location data), connects to known Wi-Fi networks, runs apps that collect and sell location data to brokers, and carries unique hardware identifiers. Push notifications themselves can leak information about your activity and can be accessed by law enforcement. Metadata, who you contacted, when, from where, and for how long, is often more revealing than content and is frequently less protected.

What it looks for: identifiers that link a device to a person, location trails, and the social graph of who communicates with whom.

Data Aggregation and Brokers

The least visible layer is also the most pervasive. Data brokers compile profiles from purchases, loyalty cards, app permissions, public records, and web tracking, then sell them. Government agencies routinely buy this commercial data, which lets them obtain information that might otherwise require a warrant. Fusion software like Palantir's then links these disparate sources into a single profile.

What it looks for: correlation. Any two data points that can be tied to the same person make the profile sharper.

Part Three: A Practical Privacy Guide

No single tool makes you invisible, and anyone selling that promise is lying. Privacy is layered risk reduction. The right approach is to think about your "threat model": what you are protecting, from whom, and how much inconvenience you will tolerate. A journalist protecting a source needs more than a person who simply does not want advertisers building a dossier. Start by deciding which category you are in, then apply the relevant layers.

The framework below moves from highest-impact, lowest-effort changes to more advanced measures.

Layer One: Communications

Use end-to-end encrypted messaging. Signal is the standard recommendation among security researchers because it encrypts message content and minimizes the metadata it retains. End-to-end encryption means only you and your recipient can read the message; the service provider cannot. This is different from transport encryption, where the provider can still see your messages in the clear on its servers.

Avoid SMS for anything sensitive. Standard text messages are not end-to-end encrypted and are easily intercepted and logged.

Reconsider your email provider. Mainstream providers can be compelled to hand over your mail. Privacy-focused providers like Proton Mail offer end-to-end encryption between users and are based in jurisdictions with stronger privacy protections. Understand that email is fundamentally hard to fully encrypt because of how the protocol works, so for truly sensitive communication, encrypted messaging beats email.

Remember that in-person conversation, with no devices present, remains the most private channel that exists.

Layer Two: Your Phone

Your phone is your biggest exposure, so it deserves the most attention.

Audit app location permissions. Set location access to "while using" or "never" for everything that does not genuinely need it. Most apps that harvest location for sale to brokers ask for "always" access they do not need.

Turn off the advertising identifier. Both iOS and Android let you reset or disable the per-device advertising ID that brokers use to link your activity across apps.

Disable Wi-Fi and Bluetooth scanning when not in use. Phones constantly probe for networks and beacons, which can be used to track location even without GPS.

Use Advanced Protection features. Both Apple and Google now offer hardened security modes (Apple's Lockdown Mode, Google's Advanced Protection) for higher-risk users.

Understand the airplane mode and power-off limits. A phone that is on is a phone that can be located. For situations where you genuinely do not want to be tracked, leaving the phone at home is more effective than any setting. A Faraday bag (a signal-blocking pouch) prevents a powered phone from transmitting, but simply not bringing the device is simpler and more reliable.

Layer Three: Web and Network

Use a privacy-respecting browser and block trackers. Firefox with strict tracking protection, or Brave, plus the uBlock Origin extension, eliminates most commercial web tracking.

Use a search engine that does not profile you, such as DuckDuckGo or Brave Search.

Understand what a VPN does and does not do. A reputable VPN hides your browsing from your internet service provider and masks your IP address from sites you visit. It does not make you anonymous, because you are simply shifting trust to the VPN provider, who can see your traffic. Choose a provider with a credible no-logs policy and an independent audit. For stronger anonymity, the Tor Browser routes your traffic through multiple relays so no single point sees both who you are and what you are doing, at the cost of speed.

Prefer HTTPS everywhere, which is now near-universal but worth confirming for sensitive sites.

Layer Four: Data Footprint and Accounts

Opt out of data brokers. Services exist to automate removal requests, or you can file them yourself with the major brokers. This is tedious and must be repeated periodically, because brokers re-acquire data, but it measurably shrinks your profile.

Minimize loyalty programs and reduce card purchases for sensitive items. Cash leaves no transaction record. Loyalty cards are explicitly designed to link your purchases to your identity.

Compartmentalize. Use different email addresses for different purposes, and consider email-masking services that generate unique forwarding addresses per site, which both reduces correlation and reveals which company leaked or sold your address.

Lock down social media. The simplest defense against facial recognition databases is to not feed them. Limit public photos, tighten privacy settings, and think before posting images that tie your face to your name and location.

Layer Five: Movement and the Physical World

This is where the Flock and camera question lives, and where honesty matters most.

The license plate on a vehicle driven on public roads is, by current US law, generally not private. Police are permitted to read plates in public, and courts have so far upheld fixed ALPR deployments. Physically obscuring, altering, or covering a license plate to defeat readers is illegal in every US state and will get you cited or arrested. This guide does not recommend it, and it is not a privacy strategy, it is a traffic offense that draws exactly the attention you were trying to avoid.

The lawful and effective measures are about reducing how much your movements are tied to you:

Drive less in tracked corridors when privacy matters, and use public transit, cycling, or walking, which generate far less identifiable vehicle data.

Be aware of where cameras cluster. ALPRs concentrate at major intersections and highway entrances. Activist and journalist communities maintain awareness of camera locations through projects like the EFF's Atlas of Surveillance, which maps known deployments by jurisdiction.

Support legal and policy change. This is genuinely the highest-leverage move available, and it is not a throwaway line. ALPR dragnets exist because city councils approve them, often with little public scrutiny. The Washington legislation introduced in 2026 to govern ALPR data retention and access is exactly the kind of measure that constrains these systems. Public records requests, city council testimony, and support for organizations like the EFF and ACLU have repeatedly forced transparency and rollbacks. Individual technical measures protect you; policy protects everyone, including the people who will never read a privacy guide.

Layer Six: Operational Habits

Tools fail when habits undercut them. A few principles tie it together:

Consistency beats sophistication. An encrypted messenger you use only sometimes protects you only sometimes. Pick sustainable practices.

Metadata is the soft underbelly. Even when content is encrypted, the pattern of who you talk to and when can reveal a great deal. Reducing the number of services that see your social graph matters.

Convenience is the enemy of privacy, and that tradeoff is real. Every default that makes life easier, saved payment methods, "always" location, single sign-on, is a default that increases exposure. You do not have to refuse all of them. You do have to choose consciously.

Assume retention is forever. Data collected today may be searchable for decades, as the Utah Data Center's design philosophy makes explicit. The most private data is the data that was never generated. Not creating a record is always stronger than trying to hide one.

Conclusion: Default Surveillance and the Possibility of Refusal

The surveillance apparatus described here was not the product of a single decision. It accreted, one program, one contract, one camera at a time, usually justified by a real concern (terrorism, crime, immigration enforcement) and usually expanding well past that justification once built. The throughline from the PATRIOT Act to XKeyscore to ImmigrationOS to the Flock camera on your corner is the steady conversion of "can be collected" into "is collected by default."

The individual measures in this guide will not return you to the privacy baseline of 1995, and no honest guide would claim otherwise. What they do is restore an element of choice. They let you decide what to generate, what to share, and what to keep off the record. They raise the cost of tracking you specifically, which matters, because mass surveillance depends on it being cheap to track everyone.

The deeper lever is collective. Encryption protects the individual; law protects the population. The history above is, among other things, a record of what public exposure can do, the Snowden disclosures forced reforms, sunset provisions, and a global encryption movement that hardened the entire consumer internet. The cameras and contracts described here are subject to the same dynamic. They depend on remaining unexamined.

So examine them. Protect yourself with the tools, and push for the rules. Both are part of the same project: the basic proposition that being left alone should be the default, and being watched should require a reason.

Further Resources

Below are direct links to the resources referenced in this article, plus additional tools and projects worth knowing about. Everything is grouped by category. All links go to the primary source.

Foundational guides

• EFF Surveillance Self-Defense — ssd.eff.org — The definitive practical guide from the Electronic Frontier Foundation. Free, regularly updated, with threat-modeling walkthroughs, tool tutorials, and scenario-based playbooks (attending a protest, protecting sources, etc). This is the single best starting point for most people.
• Privacy Guides — privacyguides.org — Independent non-profit that catalogs and vets privacy-respecting software. Their recommendations are conservative and technically strong. Good for finding the "right" tool in any given category.
• PrivacyTools — privacytools.io — Sister-style resource with a broader catalog. Useful for browsing alternatives to specific products (e.g., "what should I use instead of Gmail").
• Freedom of the Press Foundation Training — freedom.press/training — Digital security guidance oriented toward journalists and sources. If your threat model is above average, this is the level to reach for.
• ACLU Privacy and Technology — aclu.org/issues/privacy-technology — Ongoing reporting, litigation, and policy work on facial recognition, ALPRs, and government surveillance. Also see their Surveillance Technologies hub.

Mapping the cameras around you

• EFF Atlas of Surveillance — atlasofsurveillance.org — Searchable map of over 10,000 documented surveillance technology deployments by U.S. law enforcement, including drones, body cameras, ALPRs, facial recognition, and more. Search by city or agency.
• DeFlock — deflock.org and the interactive map at maps.deflock.org — Crowdsourced open-source project that maps Flock Safety and other ALPR cameras onto OpenStreetMap. As of early 2026 the underlying registry passed 336,000 tagged camera locations worldwide. Backed by the EFF. You can browse cameras near you and even generate privacy-optimized driving routes that avoid known ALPR clusters.
• EFF Street-Level Surveillance — eff.org/issues/street-level-surveillance — Explainers for each major category of local surveillance technology (ALPRs, facial recognition, drones, stingrays, gunshot detectors, etc). If you want to understand a specific device before responding to it, start here.

Anti-recognition tools and techniques

These sit at the artistic/experimental end of the spectrum. They're not silver bullets and their effectiveness varies against modern systems, but they're the honest state of the field for people who want to reduce their machine-visibility.

• CV Dazzle — adam.harvey.studio/cvdazzle — Artist Adam Harvey's foundational project on using hair styling and asymmetric makeup to defeat computer-vision face detection. The technique targets face detection specifically (the step that comes before recognition), so if the algorithm never detects your face, it also never identifies you. Effectiveness against modern deep-learning systems is more limited than against the older algorithms it originally targeted, but the concept and looks are worth understanding.
• Adam Harvey's full project index — adam.harvey.studio — Includes Stealth Wear (anti-thermal-imaging clothing), DFACE.app (open-source face redaction), Think Privacy (public awareness), and other counter-surveillance work.
• Fawkes — sandlab.cs.uchicago.edu/fawkes — Free software from the University of Chicago's SAND Lab. Makes tiny, invisible-to-humans changes to your photos before you post them, which "poisons" facial recognition training. Over 840,000 downloads. Won't help against real-time cameras, but helps prevent your online photos from being scraped into future recognition databases (the Clearview AI problem).
• Reflectacles — reflectacles.com — Commercial anti-facial-recognition glasses that use infrared-blocking lenses and reflective frames to disrupt cameras that use infrared illumination (common on nighttime security cameras and 3D face-mapping systems like Face ID). Made by an independent designer, in production since 2015.
• Computer Vision Dazzle (Wikipedia overview) — en.wikipedia.org/wiki/Computer_vision_dazzle — Broader academic and cultural context on the technique, including its use in protest movements.

Essential privacy tools

• Signal — signal.org — The gold standard for end-to-end encrypted messaging. Available on iOS, Android, Mac, Windows, and Linux.
• Tor Browser — torproject.org — For strong anonymity when browsing. Routes traffic through multiple relays so no single point sees both who you are and what you're doing.
• Tails OS — tails.net — A live operating system you boot from a USB stick. Runs Tor by default, leaves no trace on the host computer. Used by journalists working with sensitive documents.
• Proton Mail — proton.me/mail — End-to-end encrypted email based in Switzerland. Includes an encrypted calendar and drive. Free tier available.
• Tuta (formerly Tutanota) — tuta.com — Alternative encrypted email provider, based in Germany.
• DuckDuckGo — duckduckgo.com — Search engine that doesn't profile users.
• Brave Search — search.brave.com — Independent search index, also privacy-respecting.
• Firefox with Arkenfox — firefox.com plus the Arkenfox user.js — For a browser hardened against tracking and fingerprinting.
• uBlock Origin — ublockorigin.com — The best free content and tracker blocker. Firefox and Chromium browsers.
• Bitwarden — bitwarden.com — Open-source password manager. Free tier is generous.
• Have I Been Pwned — haveibeenpwned.com — Check whether your email address or phone number has appeared in known data breaches.

Data broker removal

• EFF's guide to opting out of data brokers — eff.org/wp/behind-the-one-way-mirror — Background on how the data-broker economy works and general opt-out guidance.
• YourDigitalRights.org — yourdigitalrights.org — Free tool that helps generate opt-out requests to major data brokers under GDPR and CCPA.
• Incogni — incogni.com — Paid service that automates broker opt-out requests on your behalf. Useful if you don't want to manage the process manually.

Historical and reporting resources

• The Snowden Archive at Canadian Journalists for Free Expression — snowdenarchive.cjfe.org — Searchable archive of the original leaked NSA documents.
• The Intercept — theintercept.com — Continues to publish investigative reporting on surveillance, founded partly to work through the Snowden material.
• The U.S. Press Freedom Tracker — pressfreedomtracker.us — Database of press freedom incidents in the United States, maintained by Freedom of the Press Foundation.

A final honest note on the counter-surveillance tools

The face-disguise and image-cloaking tools listed above are the state of the art for individual counter-surveillance, but their effectiveness is genuinely a moving target. Every technique here has been at least partly caught up with by more recent recognition models. The best of them still raises the cost of identifying a specific person, which matters, but no honest resource would tell you any of these makes you invisible. The most robust protection remains not being in the database in the first place, which is why the tools higher up this list (Signal, Tor, minimizing your public photo footprint) tend to give the strongest return on effort for most people. The most robust protection of all remains collective political action to constrain the systems themselves.