pgAdmin lets database users run commands on the pgAdmin computer
A flaw in pgAdmin's data export tool can let a logged-in database user sneak extra commands past its checks. Those commands run on the computer where pgAdmin is installed, not inside the database.
| Severity | CriticalCVSS 9.4 of 10 |
|---|---|
| Being exploited? | UnclearNo reports either way yet |
| Fix available? | YespgAdmin 4 9.19 or later (the fix rejects any backslash inside a double-quoted identifier) · since 30 Oct 2026 |
| How sure is this? | Official sourceBacked by an official source |
How the attack works simplified
- 1Database userHolds export permission by default
- 2Crafted export queryDouble-quoted identifier with backslash
- 3Quote check fooledParenthesis escapes the \copy wrapper
- 4psql runs TO PROGRAMExecutes on pgAdmin host
- ResultAttacker runs commands on pgAdmin host
What happened
pgAdmin 4's Import/Export Data tool wraps query exports in a psql \copy command and checks quote balance, but mishandles backslashes inside double-quoted identifiers when standard_conforming_strings is off. This lets a closing parenthesis escape the wrapper and a TO PROGRAM clause run through psql on the pgAdmin host. It is a bypass of the earlier fix for CVE-2026-17566.
Are you affected?
You are affected if you run pgAdmin 4 before version 9.19 and users have the Import/Export Data permission.
pgAdmin 4 before 9.19
What to do
- Update pgAdmin 4 to version 9.19 or later
- Until updated, restrict the Import/Export Data permission to trusted users only
- Review pgAdmin host accounts and logs for unexpected command execution
Timeline
- 30 Oct 2026, 20:17 UTCPublicly disclosed
- 30 Oct 2026, 20:17 UTCReported by NVD · NVD
- 30 Oct 2026, 20:20 UTCPicked up by Zero Radar
Sources 1 report, oldest first
Official vendor, government CERT or advisory database
- CVE-2026-116241NVD · 30 Oct 2026, 20:17 UTC
Technical details
- ID
- CVE-2026-116241
- CVSS
- 9.4 / 10
- Disclosed
- 30 Oct 2026, 20:17 UTC
- On Zero Radar
- 30 Oct 2026, 20:20 UTC (3 min after disclosure)
- Last change
- 30 Oct 2026, 20:20 UTC
The plain-language parts and the attack diagram were written by an AI model (anthropic/claude-haiku-5.5) from the sources above. Facts like versions, exploitation and dates come from the sources themselves. How Zero Radar works