We don't identify who did something; we stop what it does. Every PryVPN server closes the ports abuse relies on, limits how fast each device can open connections, and slows down or quarantines a connection that scans, brute-forces or floods, for 2 or 30 minutes, in memory only. Then it forgets. When a report arrives, a person acts on the server, and we count every report and request from authorities in our transparency report.
Anyone who runs a VPN gets emails like this: "Your IP address 203.0.113.7 attacked our server at 14:02. Who was it?" A VPN that keeps no activity logs can't answer. That's the point of it. It also raises a fair question: if you can't tell who did it, how do you stop it?
This is how we do it, with the actual numbers our servers use.
Why abuse matters to everyone on a VPN
Everyone connected to a PryVPN server shares its IP addresses. When one person sends spam or scans other people's computers through them, those addresses land on blocklists, and everyone else on that server starts seeing captchas and blocked sites. Stopping abuse isn't only about being a good neighbour on the internet; it keeps the service usable for the people who use it properly.
There are two easy ways out, and both are wrong:
- Log everything, so you can always answer "who was it?". That turns a privacy tool into a surveillance system.
- Do nothing, and let the IP addresses burn.
Our approach: judge behavior, not identity
Our servers look at how traffic behaves, never at what it contains or where you browse. The rules run inside each server's kernel, in real time, and they track a connection only by its address inside our network, not by account and not by real IP address.
What every server stops on its own
| Rule | Our default |
|---|---|
| Ports used for spam, worms and amplification attacks | Outgoing TCP 25 (email), 23 and 2323 (Telnet), 137–139 and 445 (Windows file sharing); UDP 19, 137–138, 1900 and 11211 are closed |
| New connections per device | 150 a second, bursts up to 300, and 2,000 open at once: far above normal use |
| Scanning | More than 300 different destinations in a minute |
| Brute force | More than 20 login attempts a minute against SSH, remote desktop or databases |
| Flooding | 20,000 packets a second at one target |
| Known malware and phishing domains | Always refused by our DNS, for everyone |
| Fair sharing | Bandwidth split evenly between the people on a server |
Each of the scan, brute-force and flood checks is a strike.
- A first strike slows that connection down for 2 minutes: only 5 new connections a second. Whatever is already open keeps working, so a person on a busy web page barely notices. An attack stalls.
- A second strike while throttled quarantines it for 30 minutes: only web browsing (ports 80 and 443) and our DNS, at 2 new connections a second.
- Then the server forgets. Technically, the tunnel address sits in a kernel set with a timeout, the way the nftables documentation describes: when the timer runs out, the entry disappears. Nothing is written to disk, and no account is flagged.
The per-device counters behind these checks live in the server's memory for up to a minute. The only thing that ever reaches our control systems is a running total, such as "3 connections throttled right now", without saying which. The fair sharing uses CAKE, a standard Linux queue discipline that shares bandwidth per host without needing to know who anyone is.
Keeping account farms out
The same idea applies to sign-ups. Creating an account asks your phone to do a small amount of computing work, more during waves of automated sign-ups, instead of showing a captcha from another company. Accounts that connect from an unusual number of networks in a day, a sign of large-scale sharing, get stricter connection limits; we keep only the daily number, never the networks. All of this is listed in our privacy policy.
When someone reports abuse
People and companies can report abuse through the form on our abuse page or by email. A person reads every report and acts within 24 hours for child sexual abuse material and requests from authorities, 48 hours for attacks and network abuse, and 72 hours for everything else.
What we can do:
- Act on the server named in the report: tighten its limits, block a port, or take it out of service.
- Suspend an account, when a report gives us something that identifies it, such as its email address.
- Confirm which of our servers an IP address belongs to, and say what we changed.
What we can't do: tell anyone who used an IP address at a given time, or where their traffic went. That link doesn't exist, for them or for us.
For authorities in the EU, our abuse page is also the single point of contact that the Digital Services Act asks for (Articles 11 and 12). We answer every request in writing, and every report and request is counted in our quarterly transparency report, including how much user data we handed over. That number can only be zero, because we don't have the data.
Why this fits a no-logs promise
Data protection law already points this way. The GDPR's principles of data minimisation and storage limitation (Article 5) ask for no more data than needed, kept no longer than needed. Stopping abuse by behavior needs a counter and a timer, not a history. We explain what "no logs" should mean in what no-logs means, and list what we do store in what we store and who pays.
What this doesn't solve
- Thresholds are a trade-off. Set them too low and ordinary users get slowed down; too high and some abuse slips through. We'll adjust them as real traffic shows where the line should be, and publish the defaults on the abuse page.
- Not every abuse is a flood. Slow, low-volume harassment doesn't trip a rate limit. In those cases we can only act when a report gives us something to act on, such as the email address of an account.
- Rules still apply. Our terms set out what you agree not to do with PryVPN, and what happens automatically if a connection does it.
If you'd like to use a VPN that protects its shared addresses without watching you, you can download PryVPN.
Sources
Spotted a mistake? Write to support@pryvpn.com and we'll fix it and update the date.
