CVE-2026-25996 is an escape sequence injection (CVSS 9.8 in the record) in Inspektor Gadget, the eBPF toolkit for Kubernetes and Linux. Event strings are printed without removing control characters. Upgrade to v0.49.1, or use JSON output, which already sanitized the content. The record does not list it in CISA's KEV catalog, and the advisory we read does not report exploitation.
What's affected
- Inspektor Gadget v0.49.0 and earlier
- Fixed in v0.49.1 and later
The flaw is in the default columns output mode of ig run, which is the mode used when you run it interactively. The advisory notes that json output was already sanitizing the content, so that mode avoids the problem.
How to check if you're exposed
Ask the tool:
ig version
Compare with v0.49.1. Also check where else ig is installed: jump hosts, laptops of platform engineers, and any container image that bundles it. The risk sits with the operator's terminal, so the machines where people run ig interactively matter most.
What to do now
- Upgrade ig to v0.49.1 or later on every machine where operators run it.
- Until then, run ig with JSON output when watching events from workloads you do not control.
- Treat any container you observe as untrusted input, including its process names and other strings.
- Tell the platform team that terminal output from observability tools is data from the workload and can be forged.
- Update any shared runbooks that show the columns output so they point at the patched version.
Why it matters
CWE-150 covers improper neutralization of escape, meta or control sequences. Terminals interpret special byte sequences as commands: move the cursor, change colors, rewrite what was already printed. If a tool echoes attacker controlled strings to a terminal without filtering them, the attacker can influence what the operator sees, and in some terminals more than that.
In this case the strings come from eBPF events produced by an observed container. A forged event, partly or fully, can carry escape sequences into the screen of whoever runs ig. The advisory describes various effects and does not list specific outcomes, so we leave it there.
The lesson applies to every tool that prints data from untrusted workloads: strip or escape control characters before they reach a terminal, a log viewer or a web page.
Quick check
Does using JSON output fully protect me?
The advisory says json output was already sanitizing content, so it is a reasonable stopgap. The fix is v0.49.1.
Why is the score so high for a terminal bug?
The record gives it 9.8 with network reach and high impact. Read the advisory for the project's own description of effects.
Sources
Last reviewed: 19 October 2026.
