Your Tor Is Not Anonymous

Witness protection runs like this. New name, new city, new life story. Papers, an address, a cover. Serious outfit, US DOJ bankroll, decades of craft. Still, witnesses get found. Not because the papers are shoddy. Because the person keeps walking the same walk, laughing the same laugh, ordering the same flat white at the same hour. You can change a name. Can't change a self.

Tor Browser runs on identical logic. Hands you a fresh IP, an encrypted route through three nodes, a cookie-cutter browser fingerprint. Good disguise. Hitch is that inside the disguise, there's still you. And you give yourself away in 60 seconds.

In brief: Tor hides your IP and flattens the browser fingerprint. Yet behavioral fingerprinting through mouse, scroll, and keyboard strips 78 to 85% of anonymity within a minute. CSS attacks work with JavaScript off. An ad blocker shrinks the anonymity set to 48 people. German police proved traffic timing analysis works outside the lab.

What Tor Actually Shields

Credit where it's due. Tor gets a lot right. Your IP hides behind three layers of encryption. Browser flattens its fingerprint: letterboxing rounds the window to 200 by 100 pixel steps, timezone forced to UTC, canvas reads blocked, WebRTC fully off, font set capped below 15. At Safest level, JavaScript, WebGL, and media autoplay all go dark. First-party isolation stops cookies from tailing you across sites. RFP reports CPU as 2 cores, and navigator.deviceMemory isn't even built into Firefox or Tor, so memory size never leaks.

Against dragnet snooping by ad networks, this holds. Trouble begins when the foe is willing to spend a bit more sweat.

Sixty Seconds to Naked

Georgetown and CMU in 2025 published Rethinking Fingerprinting in PoPETs: Crichton, Cranor, Christin. Two years of watching, 150,000 users. Question: what happens after a break in continuity? User wipes cookies, swaps browsers, restarts a Tor session. On paper, a new visitor. In truth, not.

Within 60 seconds of ordinary browsing, behavioral fingerprinting peeled away 78 to 85% of a user's anonymity. Out of a candidate pool of 1,000, only 150 to 220 stayed standing. After 10 minutes, anonymity loss hit 90%. How: mouse movements, scroll rhythms, typing speed, page-visit sequences. All as singular as handwriting. All of it sails through every technical barrier, because Tor standardizes the browser, not the human behind it.

Researchers' takeaway reads like satire: visit fewer pages, spread your sites around, steer clear of niche content. Put plainly: to be anonymous, be dull. Tor can cloak the pen, not the hand gripping it.

A Fortress With No Roof

Tor at Safest kills JavaScript. User feels bunker-grade safe. No JS means scripts can't fingerprint you. True, and only half the story.

CSS works. Always works. CISPA team led by Trampert et al. showed this at NDSS 2025 in Cascading Spy Sheets. Container queries and calc and font width measurement tag 1,176 browser-OS combos at 97.95% accuracy. Without one line of JavaScript. Pure CSS fires a request to a unique URL, server grabs the fingerprint.

Tor Project owned up to the flaw. Then misjudged its reach. From the paper: Tor Project acknowledged the issue but wrongly expected other font fingerprinting mitigations, namely disabling scripting and font loading, to blunt the attack. Walls up, gates shut, sky wide open.

Your Bodyguard Sells Your Schedule

Plenty of Tor users bolt on uBlock Origin or a kindred blocker. Makes sense: fewer trackers, less noise, faster loads. Except the Epfl and CISPA study Double-Edged Shield at USENIX Security 2025 found a side effect.

An ad blocker can be read by its setup. A site checks which elements are blocked, blocking pattern gives away the set of filter lists. Detection accuracy: 84% of lists. Out of 577,000 rules, checking just 45 is enough to shrink the anonymity set to a median of 48 people. 0.2% of the crowd.

Tool built for privacy turned into its biggest leak. You hired a bodyguard, and he's hawking your schedule to anyone who asks.

When Theory Became Practice

Academic papers on Tor timing analysis go back to 2005. Idea is clean: observer logs packet timing at Tor entry and at exit. If patterns match, user is tagged. For two decades, crypto community called this theoretically doable but impractical at scale.

September 2024, German Federal Criminal Police, the BKA, laid out how they unmasked the Boystown platform admin between 2019 and 2021. Method: timing analysis of traffic at Tor entry nodes and Ricochet messenger servers. Not a one-off job. Node watching stretched over several years. A Chaos Computer Club spokesperson confirmed law enforcement had repeatedly and successfully run timing analysis attacks against selected Tor users over that span.

Detail easy to miss: admin ran an old Ricochet build without Vanguards protection. Tor 0.4.7 and later added vanguards-lite. Tor Project gave its stock answer: not a new flaw, users should keep software current. Fair enough. Fact stands: a state with deep pockets unmasked a specific person on the Tor network. Not in a lab. In a live criminal case.

Death by a Thousand Cuts

No single trick breaks Tor on its own. They don't work alone either.

Start with the math. Tor has roughly 2 million daily users, 1.95 million as of October 2024. Internet has 5.5 billion people per ITU 2024. Tor = 0.035%. Just using Tor makes you about 2,800 times more conspicuous than the average user. Like wearing a balaclava in a shopping mall. Nameless among fellow balaclava wearers, but the security feeds already plucked you from the throng.

Stack the layers. CSS fingerprinting at NDSS 2025 nails browser type and OS. Ad blocker fingerprinting at USENIX 2025 narrows the set to 48 people. Behavioral fingerprinting from PoPETs 2025 strips 78 to 85% of anonymity within 60 seconds. Traffic timing analysis by the BKA from 2019 to 2021 ties entry to exit. A sweeping survey of website fingerprinting attacks at arXiv:2510.11804 from October 2025 shows deep learning models hit high accuracy for visited-site identification even in open-world setups, multi-tab attacks included via Mtwfa-SEG.

Each layer peels off a slice of anonymity. Tor guards against each attack one at a time. Nobody guards against all of them at once.

FAQ

Is Tor useless? No. Shields against mass surveillance, ad trackers, idle watchers. If your foe won't spend on timing analysis and behavioral fingerprinting, Tor works. Problems start when the foe is hunting you by name.

Does a VPN over Tor help? VPN hides from your ISP the fact you're on Tor. Doesn't block behavioral fingerprinting, CSS attacks, or timing analysis at exit nodes. Different layer, not a stand-in.

Does Safest level fix things? Partly. Killing JavaScript stops most script-driven attacks. Doesn't stop CSS fingerprinting at NDSS 2025, behavioral analysis at PoPETs 2025, or timing correlations per the BKA. Three of four attacks laid out here work with JS off.

What about Tails and Whonix? Lock down the OS and surroundings. Strips part of the fingerprint: no Adobe fonts, no oddball software. Doesn't strip behavioral fingerprinting or timing analysis. Better than Tor Browser on Windows. Not a silver bullet.

What did the PoPETs 2025 researchers recommend? Word for word: visit fewer pages, spread your sites around, steer clear of niche content. Real anonymity means being the dullest user on the wire. More set your browsing pattern, easier you are to spot.

---

Read Next

• Server-Side Fingerprinting: TLS and HTTP Bypass Tor
• Your Fonts Are Your Passport
• The WebRTC Leak: Why Tor Blocks It Completely

---

WHOX - We tell real signals from manufactured ones. Check your fingerprint at whox.com and find out what signals leak from your browser and how unique you are.