On this page (7 sections)

A critical flaw, CVE-2026-18535, has been recorded for IBM API Connect. According to the GitHub Advisory Database entry, two version ranges are affected, and the problem combines a Regular Expression Denial of Service (ReDoS) weakness with improper neutralisation of wildcard query operators. The record was published on November 7, 2026 with a CVSS score of 9.6.

Anyone who runs API Connect as an API gateway or behind a developer portal should check which version is deployed. The public record is thin, so this article keeps confirmed facts apart from open questions and focuses on how to judge your exposure and what to do next.

What the record says

The description lists two separate problems in the same product: a ReDoS weakness and improper handling of wildcard characters in queries. The NVD entry repeats the GitHub text word for word, so together they amount to one source, not two confirmations.

The advisory references two IBM support pages: node 7291753 and node 7292434. Their contents were not available to us, so we cannot say which fixes, interim fixes or mitigations IBM recommends. Treat those pages as the authoritative place for remediation guidance.

The GitHub entry is marked unreviewed and lists affected and patched versions as unknown in its package fields, since no package from a supported ecosystem is attached. The version ranges in this article come from the description text, not from structured version data.

Technical background in plain terms

ReDoS

A regular expression is a text-matching pattern used to validate or search input. Some patterns have a pathological worst case: for a carefully shaped input, the engine tries a huge number of combinations before giving up. CWE-1333 describes this as a pattern whose worst-case cost is inefficient and possibly exponential. A single request can keep a CPU core busy for a long time, and a handful of them can slow or stall a service.

Wildcard query operators

Wildcards are characters that stand for any run of text in a search or filter. If input containing them is not neutralised before reaching the query engine, a caller may change what the query matches or how expensive it is. The research does not name the API Connect component or endpoint involved, so we do not speculate about specific behavior.

Reading the CVSS vector

The GitHub advisory shows CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:C/C:H/I:N/A:H. In practice that means:

  • Network attack vector: reachable remotely.
  • Low attack complexity and no user interaction.
  • Low privileges required: some access, such as an ordinary account, is needed, but not administrator rights.
  • Changed scope: impact can extend beyond the vulnerable component.
  • High confidentiality and availability impact, no integrity impact.

A plain ReDoS usually affects availability only, so the high confidentiality rating may relate to the wildcard query issue. The sources do not explain this; that link is our own reading of the vector, not a vendor statement.

CVSS score (GitHub Advisory Database)
9.6
Inefficient Regular Expression Complexity
CWE-1333
Affected version ranges
2

Who is exposed

Any deployment running one of these ranges is in scope:

  • 10.0.8.0 through 10.0.8.12
  • 12.1.0.0 through 12.1.1.2

Because only low privileges are required, risk is highest where API Connect components can be reached from the internet or by many internal users, and where regular developer, consumer or tenant accounts exist. A deployment limited to a few trusted administrators is less exposed in practice, but with a network attack vector you should confirm that rather than assume it.

A gateway also fronts other services. If it slows or stalls, every API behind it can look down, so the business impact may reach well past the component named in the advisory.

How to check your exposure

Start by finding the exact version of each API Connect deployment, including test and staging. The method depends on how it was installed (a management interface, a Kubernetes or OpenShift deployment, or a vendor-managed service), so use IBM's administration tooling or your deployment records. On Kubernetes, the read-only command below lists container images of running pods in one namespace so you can compare them with the affected ranges. Run it where kubectl is configured and replace the placeholder with your namespace:

List container images in an API Connect namespace (read-only; run where kubectl is configured)Shell
kubectl get pods -n <your-apic-namespace> -o jsonpath='{range .items[*]}{.metadata.name}{"\t"}{range .spec.containers[*]}{.image}{" "}{end}{"\n"}{end}'

Image tags do not always match the product version, so confirm the version through IBM's own tooling or documentation before concluding anything. For a hosted or managed offering, ask the provider whether your instance falls in scope.

Then look at your logs and metrics for warning signs: CPU spikes on API Connect components, slow responses on search or filter endpoints, and repeated requests from one account or address with long, repetitive query strings. These point to abuse in general and are not proof of this particular flaw.

What to do now

  1. Inventory every IBM API Connect deployment and record its exact version.
  2. Compare each version with 10.0.8.0 to 10.0.8.12 and 12.1.0.0 to 12.1.1.2.
  3. Read both IBM pages, 7291753 and 7292434, for fixed versions or interim fixes, and follow IBM's upgrade steps.
  4. Schedule the update in a maintenance window, starting with internet-facing and multi-tenant deployments.
  5. Until patched, review who can register or obtain credentials, and remove unused or low-trust accounts that can reach the API.
  6. Apply rate limits and request size limits in front of the gateway where possible to reduce the effect of expensive requests.
  7. Set alerts on CPU, latency and error rate for API Connect components.
  8. After updating, confirm the new version is running and that your APIs behave normally.
Caution. Upgrading or patching an API gateway can interrupt traffic and restart components. Test in a non-production environment first and follow IBM's documented procedure and rollback guidance.
Tip. A web application firewall or reverse proxy with rate and size limits can soften expensive-request abuse, but it is a stopgap and not a substitute for the vendor fix. See what a WAF is for its strengths and limits.

Monitoring after the fix

Denial of service issues usually show up first as performance symptoms. Keep baselines for latency and CPU on gateway and manager components and alert on deviations. Keep access logs long enough to review the period before patching, looking for repeated heavy requests from single accounts.

From the outside, uptime and response-time checks reveal whether public API endpoints slow down or stop answering, the most visible effect of a ReDoS condition. Webforta Monitor covers uptime, certificate, DNS and header changes, and uptime monitoring explained describes the approach. External monitoring shows that something is slow, not why, so combine it with internal logs and metrics.

Open questions

  • Fixed versions: not stated in the CVE record or GitHub advisory. Check IBM's support pages.
  • Exploitation: none reported in the research and no CISA KEV listing. That is absence of evidence, not proof that attacks are not happening.
  • Affected component and endpoint: not described in the sources.
  • Independent confirmation: only the CVE record and its GitHub copy cover this so far, and the NVD page repeats the same text.

If IBM or other researchers publish more detail, re-check your exposure against it.

Sources

This article is based on the public sources below. Details can change as vendors and researchers publish more; check the original advisories before acting on version numbers.

  1. CVE-2026-18535 - GitHub Advisory Databasegithub.com
  2. CVE-2026-18535: IBM API Connect 10.0.8.0 through 10.0.8.12 and 12.1.0.0 through 12.1.1.2 are vulnerable to a Regular Expression Denial of Service (ReDoS) attack and to…nvd.nist.gov ·
  • IBM API Connect
  • ReDoS
  • CVE-2026-18535
  • API security
  • denial of service