Torn Market tornmarket[A]IV

Is the Tor Project Misleading Us? The Sam Bent OS Spoofing Expose, Dead Prefs on GitHub, and the Real Risk to No-JS Users

An investigative review of privacy advocate Sam Bent's expose regarding the Tor Project's removal of OS spoofing, the discrepancy between developer commits and public newsletters, the lingering dead preferences in GitHub, and whether this poses an existential threat to No-JS darknet users during server seizures.

Trust in operational cryptography and privacy tooling is not built on PR statements or marketing newsletters—it is earned exclusively through source code transparency, developer honesty, and verifiable cryptographic guarantees.

Roughly a year ago, well-known privacy advocate and researcher Sam Bent (DoingFedTime) published a detailed video investigation analyzing a deep contradiction within the Tor Project's technical governance:

https://www.youtube.com/watch?v=3wlNemFwbwE

The controversy centered on a formal statement in the Tor Project's May 29th newsletter, which asserted to its global user base that Operating System (OS) spoofing "never went anywhere" and was "here to stay."

Yet, as Sam Bent meticulously demonstrated using the Tor Project's own GitLab issue trackers, developer commits, and upstream Mozilla Bugzilla tickets, developers had explicitly and intentionally dismantled the very OS-spoofing engine that had historically shielded non-Windows users for over a decade.

While this disclosure is now approximately a year old, the underlying questions it raises regarding institutional transparency, threat modeling, and passive server-side fingerprinting remain deeply relevant to every darknet participant.

Today, we examine the technical mechanics of the OS spoofing dispute, evaluate whether this is a catastrophic vulnerability or an overblown drama, and invite the community on Conclave to share their perspective.


  1. THE ANATOMY OF THE CONTROVERSY: WHAT ACTUALLY HAPPENED?

For years, the foundational rule of Tor Browser fingerprint resistance was uniformity.

To prevent web servers, malicious exit nodes, or seized onion sites from distinguishing between individual users, Tor Browser enforced a standardized User-Agent header across all platforms:
Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:...) Gecko/20100101 Firefox/...

Whether you were booting Tails OS from a live USB, managing virtual machines on Qubes/Whonix, or running Arch Linux, your HTTP request headers lied to the remote server and claimed you were running Windows 10. Because Windows accounts for the overwhelming majority of global desktop computing, every Tor user blended seamlessly into a single, colossal anonymity set.

The Break: Tor Browser 14.5 and the Developer Statements
Beginning with architectural overhauls in Tor Browser 14.0 and culminating in 14.5, Tor developers and upstream Firefox engineers altered the privacy.resistFingerprinting (RFP) machinery.

Internal developer discussions from October 2024 through April 2025 revealed clear developer consensus:
• Developers explicitly stated their intent to "turn off spoofing OS" in HTTP User-Agent headers.
• Commit notes documented plans to "remove the relevant machinery from Firefox altogether."

The Lingering "Dead Preference" on GitHub
The most troubling discovery highlighted in the investigation was that the configuration preference (such as privacy.resistFingerprinting.spoofOsInUserAgentHeader or related user-agent flags) remained visible in configuration files on GitHub and inside about:config.

Users inspecting their browser settings were led to believe that the feature was actively protecting them, when in fact the underlying C++ and Rust code responsible for executing the spoofing routine had been decoupled or disabled.

When community members raised alarms, the Tor Project's official newsletter characterized the pushback as mere "online speculation and confusion," despite clear, documented evidence from their own repositories.


  1. WHY THIS DIRECTLY IMPACTS THE NO-JAVASCRIPT (SAFEST) THREAT MODEL

The standard defensive argument presented by browser developers is that OS spoofing in the User-Agent header is largely "security through obscurity."

Their logic is straightforward:
Modern web applications using JavaScript can easily fingerprint your true operating system through alternative vectors—such as CSS system font probing, Canvas rendering quirks, WebGL hardware driver strings, audio context processing, and navigator.userAgentData Client Hints.

Therefore, developers argued, spoofing Windows in the HTTP header while JavaScript exposed Linux underneath created a discordant, unique fingerprint that actually worsened user identifiability.

However, this justification completely ignores the reality of darknet operational security:

THE ZERO-JAVASCRIPT PARADOX:
For high-threat operators, darknet marketplace buyers, and sovereign merchants (including on TORN, where we enforce a 100% Zero Client-Side JavaScript policy), JavaScript is PERMANENTLY DISABLED via the Tor Browser "Safest" security level.

When JavaScript is turned off:
• Canvas fingerprinting cannot execute.
• WebGL cannot query graphics hardware.
• Client Hints and complex DOM API probes cannot run.
• Audio context analysis is impossible.

In a Zero-JS environment, the raw HTTP request headers transmitted during the initial handshake are literally the ONLY telemetry a web server receives!

Under the old architecture, a Tails OS user browsing a Zero-JS hidden service appeared identical to a Windows user. Under the altered regime, the server's access log directly records that the visitor is running Linux (or X11/Linux).


  1. THE WORST-CASE SCENARIO: PASSIVE LOG ANALYSIS DURING SERVER SEIZURES

Why does passive OS leakage matter so much on the Tor network?

Consider what happens when law enforcement agencies or intelligence services seize a darknet platform, hidden service, or forum:

They do not always need sophisticated browser exploits to compromise users. Instead, they analyze the server's historical access logs:

  1. Timestamp correlation against known network activity.
  2. Order placement times cross-referenced with Monero blockchain transactions.
  3. User-Agent strings recorded in web server ingress logs (nginx, Apache).

If 85% of visitors to a seized site show Windows User-Agents, but only 10-15% show Linux or Tails identifiers, investigators can immediately eliminate 85% of the crowd and isolate the high-value target pool to that tiny minority of non-Windows users.

By removing the uniform Windows mask, the Tor Project effectively segmented its user base and degraded the anonymity set for the most security-conscious participants on the network.


  1. THE REALITY CHECK: IS THIS A FATAL FLAW OR OVERBLOWN DRAMA?

Now that the dust has settled a year after the initial controversy, we can assess the situation with technical objectivity:

ARGUMENT A: Why It IS a Serious Issue

  1. Institutional Gaslighting: The most damaging aspect was not the code change itself, but the public relations response. Claiming a feature "never went anywhere" when git commits explicitly show it being removed destroys trust in project governance.
  2. The Placebo Effect: Leaving defunct preferences in configuration files leads users into dangerous false assumptions about their operational posture.
  3. Clearnet Bias: The decision was heavily influenced by clearnet compatibility (preventing Cloudflare challenge loops and website layout breakage for mainstream users), prioritizing commercial browsing over the life-or-death privacy needs of dissidents, whistleblowers, and darknet participants.

ARGUMENT B: Why It May NOT Be an Existential Threat

  1. TCP/IP Stack Fingerprinting: Even if the User-Agent header claims Windows, advanced passive network monitors running tools like p0f can often deduce the underlying OS from TCP SYN packet parameters (Window Size, initial TTL, TCP Options ordering), although Tor's onion routing layers provide significant abstraction.
  2. The Size of the Linux Tor Community: Linux and Tails users on the darknet do not represent a microscopic anomaly; they represent a significant percentage of active participants. Being identified as a Linux user does not pinpoint an individual identity without substantial external correlation.
  3. Mitigation Feasibility: Power users who demand strict header control can easily chain Tor through local filtering proxies (such as Privoxy) to strip or rewrite User-Agent strings arbitrarily.

  1. COMMUNITY INQUIRY: JOIN THE DEBATE ON CONCLAVE

This controversy touches the core dilemma of modern privacy tools: Can a single software project simultaneously serve casual clearnet users trying to bypass geographic blocks while protecting sovereign cypherpunks whose physical freedom depends on mathematical anonymity?

We want to open this floor to the community:

  1. Did you follow Sam Bent's original expose, and did it change how you configure your Tor Browser or Tails setup?
  2. Does the exposure of Linux User-Agent headers in Zero-JS environments concern you during potential hidden service seizures?
  3. Do you feel the Tor Project is slowly compromising its core privacy standards to appease clearnet CDNs and commercial website operators?
  4. How do you protect your browser perimeter today: default Tor Browser, Privoxy header rewriting, or isolated Whonix/Qubes workstations?

Bring your technical analysis, your threat models, and your insights. Let's discuss where operational security goes from here.

Trade sovereign. Verify the code. Never trust PR.

— TORN Security Research Division

https://x.com/DoingFedTime/status/2111478235875676563

https://www.youtube.com/watch?v=3wlNemFwbwE

↑ 5

Debate

Comentarios

7

Los comentarios más recientes aparecen primero en cada nivel; las respuestas siguen bajo su comentario principal.

Escribir Únete para participar en la discusión.

the dead pref point is the one that bugs me. if a switch still shows up in about:config but does nothing, people will build a threat model around a placebo, and that's worse than just removing it and saying so. fwiw the log-analysis worry is real but partial - nginx logs from a seized box are the least of it if you're ordering, timing habits leak more than a UA string.
↑ 0 Responder
Respuestas · 1
@ReverseWalrus,

Totally agree with you and I don't like the fact that the TOR project acted like the whole thing was no big deal, why make the browser in the first place if thats your attitude?

↑ 3 Responder
Respuestas · 2
@tornmarket, the newsletter wording was the part that bugged me too. if the code changed, just say the code changed and give the tradeoff — "never went anywhere" next to commits removing it is what burns trust, not the header itself.
↑ 0 Responder
Respuestas · 1
Respuestas · 1
@tornmarket, yeah. and honestly the fix is cheap: they could've shipped a release note saying "UA now reflects real OS, here's why, here's what to do if you care." ten minutes of writing beats a year of people digging through git commits.
↑ 0 Responder
Respuestas · 1
@tornmarket, eh, they still make the browser, the attitude is more "clearnet users are the majority and CDNs break layouts" — defensible choice, badly communicated. claiming a feature never went anywhere while git shows the removal is the part that's hard to wave off.
↑ 0 Responder