Tor traffic detection is often framed as a hard problem, but the network is honest about itself: Tor publishes its exit relays, so the IP that reaches your server can be checked against a public list. The hard part starts after the flag appears. The same exit node can carry a journalist, a privacy-conscious customer, a researcher and someone opening a fourth free trial in one afternoon. A policy that treats them all the same blocks paying users while the abuse moves to residential proxies.
This guide covers both halves: how exit-node detection works, which signals to combine with it, and how to write a policy that separates privacy users from abusers on the actions that cost you money.
The reference product is Prynt Identify, which returns Tor, VPN, proxy and datacenter signals next to device and behavior data, but the decisions below apply to any stack.
What Tor traffic looks like in your logs
The client picks three relays and only the last one — the exit — connects to you, so your logs record the exit IP, never the visitor’s. The Tor Project publishes the current exit set as a consensus document anyone can fetch. Detection is a lookup, not a guess: keep the list fresh, check the client IP when the request arrives, and store the verdict with the request instead of re-deriving it later.
The list covers most traffic, not all
Onion-service traffic never leaves through an exit node, and some relays refuse to exit to arbitrary ports. Expect a small share of Tor requests the list does not match. That gap is fine for policy; it is not fine for a compliance claim, so do not write “we detect all Tor” in your documentation.
An exit IP is not a verdict
Thousands of people share the same exit address at the same moment. Blocking the address blocks all of them. Using it as one input among several keeps the signal and drops the collateral damage.
Detecting exit nodes without guessing

Check the routes that matter
Run the lookup on paths that create accounts, sessions or charges: signup, login, trial start, checkout, OTP requests. Running it on static assets adds cost and tells you nothing new.
Pair the network signal with device and behavior
An exit IP says how the request arrived. It does not say whether the device has already opened five accounts this week or whether the client is a real browser. Device fingerprinting answers the first question; bot and AI-agent detection answers the second. Prynt Identify returns both alongside Tor, VPN, proxy and datacenter signals, a stable visitorId, tampering detection and shared reputation, so the decision rests on several signals instead of one shared address.
Writing a policy that separates privacy users from abusers
Judge the action, not the network
- Browsing pages, comparing plans, reading docs: allow. There is nothing to protect.
- Signup: allow, then step up to email verification when other signals agree.
- Free trial and bonus credit: step up or cap by device rather than by IP. Bonuses and referral credit are what abusers farm.
- Login: step up with a passkey or an emailed code, and watch the session for credential-stuffing patterns.
- SMS and OTP: cap per device, per number and per network. Repeated OTP requests through Tor look like pumping, not privacy.
- Checkout: allow, and step up when the billing country, card and device disagree. Payment risk stays with your processor.
Prefer step-up to block
A challenge costs an abuser a resource they would rather not spend, and an honest user ten seconds. A block costs you the customer and the support ticket that follows. Save blocks for combinations where intent is clear.
Watch it before you enforce it
Ship the policy in log-only mode first. Prynt Control runs a mirror period where every decision arrives with a reason code and nothing is blocked. Compare the flagged cohort against chargebacks, trial-to-paid conversion and support tickets. If Tor-flagged users convert at the same rate as everyone else, your policy should say so, in writing, before anyone reaches for a block.

What to store with each decision
Keep the request ID, the visitorId, the network verdict (Tor, VPN, proxy, datacenter), the reason codes and the action you took. Store hashes rather than raw identifiers, and set a retention window you can explain. When a customer asks why they were challenged, that record is the answer; when the abuse pattern shifts, it is the evidence you need to change the policy.
FAQ

Does blocking Tor stop abuse?
Rarely. It removes one cheap identity source and pushes abusers toward residential proxies, while removing every privacy-conscious customer who might have paid you. Detection is worth building; blanket blocking is a business decision to make deliberately.
Is it legal to accept Tor traffic?
Accepting Tor traffic is legal in most markets; blocking it is a business choice that belongs in your terms of service. Treat the IP as personal data under privacy laws such as the GDPR, with a lawful basis and a retention limit. This is not legal advice.
How do I tell a privacy user from a scraper on Tor?
Look at the session, not the exit address: pages per minute, whether JavaScript renders, keyboard and pointer activity, account age, and whether that device has done this before. A scraper sweeps many URLs quickly; a person reads one and leaves.
Try Prynt Identify on your own traffic
Start with the free plan, watch the decisions in log-only mode, and turn on enforcement when the numbers agree with you. pryntid.com
Try it free
Prynt is device intelligence with a free plan: 5,000 identifications a month, visitor IDs with confidence and bot detection — no credit card. Upgrade to Pro for VPN, proxy, tampering and behavioral signals.