Torzon Tor Browser Bundle Hygiene in 2026: Patching, Profiles and Per-Session Isolation

The security slider in Tor Browser still doesn’t do what most users think it does. Some of the prefs it toggles require a full restart to take effect. Others apply immediately. The result: your actual fingerprint depends on what state the browser was in when you last booted it, not on where the slider sits right now. The Tor Project confirmed this in May 2025 and tracked it under issue #42572. As of the 15.0 release cycle, they are exploring a restart prompt and a redesign that aligns the slider with their updated threat model, possibly delegating more of the back-end to NoScript. Until that ships, you are partially fingerprintable in ways the UI does not disclose to you. That is not a minor footnote. It is the entire reason your “High” security level might be doing less than you assume.

What “Patching” Means on an Amnesic Bundle

If you are running Tor Browser inside Tails or an equivalent live environment, the word “patch” has a different rhythm than on a desktop install. Tails runs entirely from the USB stick in RAM and wipes itself on shutdown. There is no background updater nagging you at 3 a.m. You get the next version the next time you write a fresh image to the USB. The practical cadence: check the Tor Project release notes, burn the new ISO, reflash, reboot. No incremental patches. No “update available” dialog you can ignore.

This also means your “patch level” is binary. You are either on the current release or you are on whatever was in the image you last flashed. There is no partial state. The downside is that you may be running a browser build that is days or weeks old between ISO writes. For most OPSEC purposes that is acceptable. For the WebTransport class of leak, it matters.

The WebTransport and WebKit Leak (March 2026)

WebTransport went live in iOS 26.4 in March 2026 and carried a DNS-prefetch and WebAuthn exfiltration path that researchers at Psylo documented with a proof-of-concept at leaks.psylo.app. Onion Browser users on its “Silver” security level were already protected because that mode enables WebKit Lockdown Mode, which disables the feature outright. Psylo patched it in 1.3.1 by blocking DNS-prefetch hints and disabling WebAuthn and WebTransport by default, with an opt-in toggle per browsing silo if you genuinely need one of those features.

The Tor Browser desktop bundle does not have a “Silver” mode. It has the slider, and as noted above, the slider is not a reliable control plane. If your research workflow touches any site that embeds WebTransport-capable content, you want to confirm which Tor Browser build you are on and whether that build’s WebKit revision predates or postdates the mitigation. The 15.0 cycle improvements to the slider should make this easier to audit, but until then, the only safe answer is: do not load untrusted content in a browser whose WebKit version you cannot verify against the CVE timeline.

Per-Session Isolation: Whonix, Qubes, and What I Actually Run

Tails gives you amnesia. One session, one boot, shutdown wipes the disk. Clean, but it is a single flat environment. If your malware sandbox test and your PGP key generation happen in the same Tails session, a compromised app can see the key material in memory. That is the trade-off you accept for convenience.

Whonix breaks that flatness. Two VirtualBox VMs: a gateway that routes everything through Tor, and a workstation that can only talk to the gateway. The workstation never sees your real IP. The gateway never sees your destination. If the workstation is pwned, the attacker has to go through the gateway to reach the outside, and the gateway does not know which internal host sent what. Application separation via the VM boundary means a malicious JavaScript payload in one Whonix app cannot reach another. The catch: Whonix does not virtualize well inside a VM. Testing it is painful. You commit to a bare-metal or single-VM install. The unsupported Live CD on the download page will not save you; it either works or it does not, and there is no “try before you buy” safety net.

QubesOS takes the model further. Each app gets its own Xen microVM. Tor traffic funnels through a dedicated Whonix gateway VM. The isolation is real, but the performance tax is real too, and the same all-or-nothing install problem applies.

For my own workflow, I run Whonix. Two VMs, no Qubes. I flash the gateway ISO, start the workstation, load Tor Browser, and treat every session as single-purpose. Need to do a PGP operation? Separate VM instance or a fresh boot. Need to browse a market and then run a vulnerability scan? Two different workstation instances, ideally two different circuit identities. The Whonix stream isolation feature ensures that two connections from the same workstation go through different exit nodes unless you explicitly bind them, which I almost never do.

Circuit Hygiene and the “New Identity” Button

Tor Browser rotates its circuit roughly every 10 minutes automatically. That is the floor, not the ceiling. If you are doing a multi-hour research session and you hit a site that logs visitor fingerprints, a 10-minute rotation does not protect you. You may have already been observed on circuit A, and circuit B three minutes later is still traceable to the same guard node.

The “New Identity” button in the Tor menu closes all tabs, clears cookies and history, requests new circuits for every connection, and restarts the browser. It is the cleanest hard reset available inside the bundle. I use it between any two unrelated research tasks. Not between pages on the same market. Between markets. Between a market session and a PGP key verification. Between a vulnerability scan and a normal browsing pass. Treat it like closing and reopening a window in the real world: once you have walked through a door, you are a different person on the other side.

What New Identity does not do: it does not clear your Whonix gateway state. The gateway still has its own circuit history. If you need a fully new anonymity layer, you restart the gateway VM. That is the nuclear option. I reserve it for when I suspect a specific session has been correlated or when I am switching between two unrelated research threads that share no cover story.

A Short, Uncomfortable Checklist

• Before every boot: Confirm your Tor Browser build number against the current Tor Project release. If it is more than one minor version behind, reflash the ISO. No exceptions for “I’ll do it next time.”
• After the slider change: If you touched the security level, restart the browser. Full restart. Not a window close. Not a “New Identity.” A process kill and relaunch. Until the 15.0 slider redesign ships, this is the only way to guarantee the new prefs actually applied.
• Per session: One purpose. One Whonix workstation instance. If the purpose shifts, kill the instance, spin up a new one, and let the gateway pick a fresh circuit.
• WebTransport / WebAuthn content: Do not load it unless you have confirmed your WebKit revision handles it. On Onion Browser, Silver mode covers it. On desktop Tor Browser, the slider is unreliable. Check the build notes.
• Exit node risk: Remember the last hop is a volunteer. Most are fine. A minority will monitor or tamper with unencrypted traffic. If the destination is HTTPS, the risk is low. If it is not, assume the exit node saw your plaintext.

What I Skip, and Why

I do not use Tails for ongoing research. The amnesia is great for a one-off meeting. It is not great for a workflow where I generate a PGP key on Monday, use it Thursday, and verify a signature on a Saturday. The key lives in a Tails persistent partition or I write it to a USB stick, and at that point the “amnesic” advantage is gone because I have re-introduced the host filesystem as a trust boundary. For sustained work, Whonix with a persistent encrypted storage VM is cleaner. You still get session isolation. You do not lose your keys between boots.

I also do not run Tor Browser as a standalone desktop app on a daily OS. The moment your browser shares a network stack with a non-Tor interface, the isolation model degrades. A leaky kernel module, a misconfigured iptables rule, a DNS query that goes out the cleartext interface: any one of those breaks the anonymity set. The VM boundary in Whonix or Qubes is what actually guarantees all egress goes through Tor. A desktop install gives you that guarantee only as long as you manually configure it, and most people do not.

The bundle is the least interesting part of the stack. The browser is a well-hardened artifact. The slider problem is a UI design flaw, not a browser flaw. The real risk is in the layer between the browser and the network: who terminates the circuit, whether your workstation can talk to anything other than the gateway, and whether you have a single-purpose session or you are juggling five tabs from four different research threads in one Whonix instance. Fix the layer below the browser first. The browser will be fine.