Clarify homepage claim about local observers and traffic analysis

The current wording on the Tor Project website is misleading regarding traffic analysis by local observers.

The homepage currently states:

“Tor Browser prevents someone watching your connection from knowing what websites you visit. All anyone monitoring your browsing habits can see is that you’re using Tor.”

Taken literally, this statement is too absolute.

I agree that Tor Browser prevents a local observer from directly seeing the destination website in the ordinary way. Such an observer may include for example an ISP, public hotspot, Wi-Fi operator, or a compromised network device. In general, this observer cannot simply inspect DNS queries, SNI values, or final destination IP addresses and read “the user visited example.com”.

However, it is not accurate to say that all they can see is that the user is using Tor.

Although a local observer cannot decrypt Tor traffic or directly identify the visited website in plaintext, they can still observe metadata from the encrypted connection between the client and the Tor network. This traffic can be modeled as a time series and analyzed through its evolving structure.

Observable features may include packet timing, inter-arrival times, packet direction, packet counts, observable packet and byte lengths at the IP/TCP/TLS layer, burst frequency, burst size, burst duration, burst ordering, burst spacing, upload/download alternation, upload/download ratios, traffic rate, throughput changes, cumulative bytes over time, total observed volume, connection start and end times, session duration, idle periods, pauses, traffic spikes, ramp-up and ramp-down behavior, repeated patterns, periodic activity, and changes in the overall shape of the traffic flow.

The observer may also analyze higher-level temporal motifs formed by sequences of bursts and gaps, such as page-load-like, download-like, streaming-like, messaging-like, login-like, search-like, or other activity-like patterns.

These features can be compared across packets, flows, connections, circuits, and sessions, or matched against previously collected traces, statistical models, or known website fingerprints.

With statistical classification or machine-learning-based website fingerprinting, a local observer may be able to make probabilistic inferences about the user’s activity. In some scenarios, those inferences may have enough confidence to be operationally meaningful.

The issue is even more serious in environments where the local observer is not just a passive network provider, but a controlled or compromised network device used as a monitoring point, traffic collection point, or exfiltration point. Such a device may be able to gather long-term traffic traces, correlate activity across sessions, compare traffic shapes against known patterns, and infer sensitive information about user behavior and confidential communication.

The current wording on the Tor Project website may give users the impression that a local observer can learn nothing beyond the fact that Tor is being used. A more accurate version would preserve the main message while avoiding an absolute claim.

Suggested replacement:

“Tor Browser prevents someone watching your connection from directly seeing which websites you visit. However, they can see that you are using Tor, and techniques such as traffic analysis or website fingerprinting may still allow probabilistic inferences about your browsing activity.”

This would be more accurate and would better reflect Tor’s actual threat model. Tor is an important and valuable privacy tool, but users should not be told, explicitly or implicitly, that a local observer can learn nothing beyond the fact that Tor is being used. That is not accurate, especially in the context of modern traffic analysis and machine-learning-based website fingerprinting.

Related point:

The same clarification may also be useful in user-facing privacy explanations, onboarding screens, or Tor Browser interface text. For example, slogans such as “Explore. Privately.” are acceptable as slogans, but users should also be clearly informed somewhere nearby that Tor improves privacy but does not make all browsing activity indistinguishable or immune to traffic analysis.

A possible user-facing clarification could be:

“Tor improves privacy, but it does not make all browsing activity indistinguishable. A local observer may still attempt to infer your activity through traffic analysis.”

This clarification would help avoid giving ordinary users a false sense of security while still accurately explaining the important privacy protection that Tor provides.