Every so often a new report claims that Tor deanonymization by law enforcement has become routine, or that a library running a Tor relay has drawn police attention. Both topics touch the same underlying question: what can Tor actually promise, and where does that promise end? The honest answer is more nuanced than either the panic or the marketing suggests.

What "deanonymization" actually means

Deanonymization means linking a specific Tor user to a specific real-world identity. It does not mean simply seeing that Tor is being used. Your internet provider can usually tell you are connecting to Tor. What Tor is designed to hide is where that traffic goes next, and who is behind it.

Tor does this by routing your connection through several relays run by different volunteers, so no single relay operator can see both who you are and what you are doing. That is a meaningful, well-documented protection, though not a guarantee that nobody, ever, can connect the dots.

A claim that did not hold up

In 2013, intelligence agencies were reported to have claimed that 80 percent of Tor users could be deanonymized within six months. That prediction did not come true. Years later it still had not happened, which is a useful data point against treating worst-case claims about Tor as settled fact.

The Tor Project has been candid that this cuts both ways: nobody, including the Tor Project, can promise "100 percent guaranteed" anonymity. Services that claim otherwise are usually overselling what they can actually deliver.

How real deanonymization cases actually happened

The clearest documented example is the attack referenced in the U.S. v. Farrell case, widely reported to involve researchers at Carnegie Mellon University who were paid by law enforcement. Relays run in that attack combined guard and exit positions with a bug in how RELAY_EARLY cells were passed along, creating an extra covert signal beyond ordinary timing data.

The Tor Project's statement on that case was direct: this approach was possible, but appeared impractical without that specific bug. Without an extra channel like that, an attacker controlling nodes is left doing complex statistical analysis that may or may not produce a usable result.

The FBI reportedly was not able to locate or identify the Tor user behind a 2016 email hack tied to a Clinton campaign staffer, even by September of that year. That is a reminder that Tor use itself does not automatically collapse under investigation.

The more common failure mode is simpler: user behavior. Downloading a file through Tor Browser and then opening it outside Tor, logging into a personal account, or reusing an identity across sessions can undo Tor's protections regardless of how strong the network itself is.

Poisoned relays and fingerprinting: what is realistic

Running malicious or "poisoned" relays remains one of the more plausible tactics available to a well-resourced adversary. Distributing trust across independently run relays is exactly why Tor's design tries to make this expensive and uncertain rather than impossible.

Browser fingerprinting is often raised as a shortcut around Tor's protections. Passive fingerprinting based on quietly observable browser characteristics is limited in what it can achieve. Active fingerprinting techniques, such as scripts probing hardware or performance details, could work in theory, but doing so silently against large numbers of people without detection is far harder than targeting a single person.

Guards, probability, and honest limits

Tor Project engineers have been unusually clear that anonymity here is probabilistic, not absolute. Guard relay selection, for example, is discussed in terms of "an x percent chance over y months" that an attacker's relay ends up in your guard set, not a flat yes-or-no guarantee.

That framing can sound unsatisfying compared to services that promise "100 percent guaranteed" anonymity. But those absolute promises usually come from centralized providers who can see everything you do on their own servers and could hand it over. Tor's distributed design is a different tradeoff, not a magic fix.

Library relays and law enforcement pushback

Public libraries running Tor relays as a community privacy service, and facing pushback from local law enforcement, has been publicly reported in a handful of cases over the years. We do not have verified figures on how common this is.

What matters is understanding the underlying confusion those situations often reflect: running a relay means forwarding encrypted traffic as one hop among several, not hosting or controlling the content of that traffic. A library relay operator generally cannot see, and is not choosing, what any individual user does on the network.

That distinction matters for policy conversations, but it does not change the core privacy math for an individual reader. Whether a relay is run by a library, a university, or an individual volunteer, no single relay operator holds both your identity and your destination at once by design.

What this means if you use Tor

Tor Browser, built on the Tor network maintained by the Tor Project, remains one of the most tested tools available for resisting local traffic analysis, whether that local observer is your ISP, your workplace network, or someone on shared wifi.

It is not a cloak of invisibility. Logging into a personal email account, reusing a username, or downloading and opening files outside the Tor environment can undo its protections.

The deanonymization record so far shows a network that has held up better than early worst-case predictions, while still relying on user discipline and honest, probabilistic claims rather than absolute ones. That balance is the actual promise Tor makes, and it is worth understanding on its own terms.