We run isIPsafe, a community database of IP addresses caught attacking servers. People who run servers report what hits them, and the reports add up to a score from 0 to 100 for each address. Most mistakes with reputation data come from reading that number as more than it is, so this post is about what the score can and can't tell you.
A score measures agreement, not guilt
The score says how sure the reporting community is right now. It is not a verdict on who is behind an address. A shared VPN exit or a mobile carrier's NAT can collect reports and still carry plenty of legitimate users.
A score of 0 is just as limited: it only means nobody has reported the address recently, not that the address is proven safe.
Recent reports count, old ones fade
Only the last 30 days matter, and a report counts for less as it ages. That is deliberate: home broadband and mobile addresses change hands, and a reputation shouldn't follow an address around forever. It also means a list you downloaded weeks ago is acting on evidence that has gone stale.
Independent sources beat volume
Each distinct reporter counts once, however many times they report the same address, and reports are weighted by how reliable the source looks. The total climbs quickly at first and then flattens, so a score of 75 or more takes several independent sources agreeing, not one noisy server repeating itself.
One report shouldn't blacklist the neighbours
A report against a whole network counts for less than a report against a single address, and it is spread across the network: the bigger the network, the less each address inside it gets. Network reports stop at a /24 for IPv4 and a /56 for IPv6. For anything larger, report the attacking addresses one by one. A small, clearly abusive block still shows up, but one report can't take thousands of innocent neighbours down with it.
Using a score without hurting real users
This part is our opinion, not a rule:
- Match the response to the cost of being wrong. On an SSH port or an admin login, blocking at a high score usually costs little. On a public checkout page, a false positive costs you a customer, so consider rate limiting or an extra verification step for scores in the middle.
- Keep an allowlist for your own addresses, partners and monitoring.
- Refresh and expire. Re-fetch on a schedule and let entries drop out, instead of pasting a one-off list into a firewall.
- Log what you block, so "I can't get in" can be answered with an address and a time.
What the data can't tell you
It can't tell you intent. An address may belong to an attacker, to a compromised machine, or to a shared connection with many users, and the reports alone don't say which. It can't tell you that an attack is still happening right now. And location data is approximate: a city only says roughly where a network is, not where someone is sitting.
If an address is listed by mistake
Check whether the reports are right first. If your own server was compromised, fix it and the score fades by itself within a few weeks as no new reports arrive. If the reports are false or malicious, ask for a delisting with the address and the reason, and a person will look at it. A reporter can also withdraw a mistaken report, and the score is recalculated straight away.
Your turn
What do you do with an address that lands in the middle of the range: block it, challenge it, or just log it? We'd like to hear what has worked for you.
How the score is calculated: https://isipsafe.com/faq#score
Top comments (1)
Is this free? can i integrate to my webapp ?