Sunday, October 11, 2026

An IP address is a lead, not a verdict: building IP investigation into Cyber Threat Analyzer

The alert with an address nobody recognizes

The fastest way to make a bad call in incident response is to treat an IP address as an answer. An alert fires, a process on a workstation has opened a connection to an address nobody recognizes, and the temptation is to paste it into a lookup site, see a red flag, and block it or isolate the host.

Sometimes that is right. Often it is not. The address may belong to a cloud provider that hosts thousands of unrelated customers, or to a VPN exit used by an employee on a trip. It may even be clean on every reputation feed because the attacker is using a compromised legitimate server. And if you isolate the machine before saving what it was doing, you may destroy the evidence that would have told you which of those it was.

In the latest version of Cyber Threat Analyzer (CTA) we built IP Investigation to support the careful version of this workflow: one that treats an address as a lead to verify, not a verdict to act on.

Six steps, in order

Experienced analysts tend to follow the same sequence, even when nobody wrote it down. IP Investigation follows it too:

  1. Extract the remote IP from the endpoint, firewall, or network logs.
  2. Determine whether it is a public IP, a private IP, or a local-network address.
  3. Look up its country, ASN, ISP, and hosting organization.
  4. Check its reputation and threat-intelligence records.
  5. Correlate the connection time with the process, user, DNS queries, and other events.
  6. Preserve the evidence before blocking or isolating anything.

The order matters. Classification comes before any lookup so that private addresses never leave your network. Preservation comes last in the sequence but first in the response, because it has to happen before anything is blocked, killed, or isolated.

Extract, classify, enrich

The remote address can show up in many forms, so CTA reads it from Sysmon network events, the endpoint agent's own network telemetry, Windows Filtering Platform events (5156 to 5158), and syslog-style firewall lines. For inbound connections it treats the source as the remote side, so you investigate the right address either way.

Classification is where privacy and noise are controlled. Every address is sorted into one of these groups: public, private LAN, loopback, link-local, carrier-grade NAT, multicast, or reserved (including documentation ranges that should never appear on the real internet). Only public addresses are ever sent to an outside service. A connection to a file server on 192.168.x.x is judged on its process and user, not on a web lookup.

For public addresses, CTA then adds country, city, ASN, ISP, and organization, plus three flags: hosting or cloud network, proxy or VPN, and mobile carrier. Results are cached, so repeated alerts about the same address do not repeat the lookup.

Reputation, then context

Reputation checks come from two places. Your own IOC denylist is always consulted, because it reflects what your analysts already know. AbuseIPDB is queried only when an analyst asks for it and a key is configured, so a busy alert queue never burns through an API quota.

The more useful step is correlation. Around the time of the connection, on the same host, CTA looks for:

  • the process that made the connection, and the user it ran as
  • DNS queries that resolved to the address, or were made by the same process
  • other connections made by the same process
  • the event that started the process, and logons by the same user
  • whether this host has ever contacted the address before

A connection to an unfamiliar address is weak evidence. The same connection from a program running out of a Temp folder, with no DNS lookup, started minutes after an unusual logon, tells a very different story.

Findings, not verdicts

CTA never labels an address "malicious" because of a lookup alone. Instead it produces a list of findings, and each one is tagged as either evidence or context. Only evidence can raise the confidence level.

Finding

Type

Weight

On your IOC denylist (high or critical)

Evidence

70

High AbuseIPDB abuse score (75 or more)

Evidence

40

Connection from a program in a user-writable folder

Evidence

15

Connection from a built-in tool often abused (PowerShell, mshta, certutil)

Evidence

15

Proxy or VPN address

Context

10

Hosting or cloud network

Context

5

First contact with this address from this host

Context

5

Evidence weight of 70 or more reads as high, 40 or more as medium, and 15 or more as low. If there is only context, the best possible outcome is "context only". A hosting or VPN flag on its own can never push an alert above low, which is the whole point: those flags explain an address, they do not implicate it.

A clean result is worded carefully too. "No indicators in the sources checked" is not the same as "safe", and the report lists every source that was not consulted so a gap is never mistaken for a clean bill. Each report also ends with a reminder that shared hosting, VPNs, cloud services, and compromised devices all complicate attribution.

Save the evidence before you act

Isolating a host, terminating a process, or quarantining a file all change the thing you are investigating. CTA can now save a snapshot of the IP intelligence for an alert first: the address, its classification, the lookup results, the findings, and the correlated events. Each snapshot is stored as a record that cannot be edited afterwards, along with a SHA-256 hash. When it is read back, the hash is checked, and if the stored copy no longer matches, the contents are withheld with a warning.

The rule is enforced where the response commands are issued, and it has three modes:

  • Off: no check.
  • Warn (the default): the action goes ahead, but a warning is written to the response audit log.
  • Require: the action is refused until the evidence is saved. In the desktop console the analyst is asked whether to save it now and continue.

Real incidents sometimes cannot wait, for example when ransomware is actively encrypting files. In any mode an operator can proceed by giving a written reason, and that reason is recorded in the audit log.

Privacy, offline mode, and what it cannot do

Some customers cannot send the addresses they investigate to a third party at all, public offices among them. For them CTA has an offline mode. In offline mode nothing leaves the network, not even the AbuseIPDB check. Country and ASN come from local MaxMind databases if you provide them, and your IOC denylist keeps working. The trade-off is honest: the free local databases have no ISP, hosting, or proxy flags, so those fields stay unknown instead of being guessed.

A few limits are worth stating plainly:

  • The first-seen and DNS findings are only as good as the logs collected. Sysmon network events are off by default, so thin coverage makes those findings weak, and the report says so.
  • An address behind a CDN or VPN describes the provider, not the person behind it.
  • The investigation covers one address at a time. It does not yet follow an attacker from the first connection through later activity on the host.
  • The free online lookup service allows 45 requests a minute and is not encrypted in transit, so production deployments should use offline mode or a paid provider.
  • Snapshot hashing detects accidental or casual edits. It does not stop someone with full database access, so ship snapshots off the machine if you need that guarantee.

What comes next

IP Investigation is available now in the Network Investigation tab of the investigation workspace, through the REST API, and in the response workflow. It works best when your endpoints send network telemetry and DNS logging is enabled, so it is worth checking that coverage before you rely on the correlation findings.

The next step we are looking at is connecting the dots across the whole incident: following a suspicious connection back to the process that made it, then forward to the files, persistence, and other hosts it touched. An address is where an investigation starts, not where it ends.

No comments:

Post a Comment

Hands-On with CTA ASM/EASM: Practical Techniques, Tips and Tricks for Effective Attack Surface Management

  A practical field guide for SOC analysts, security engineers, vulnerability managers, penetration testers, and cybersecurity consultants I...