Zero Radar
← feed
HighVulnerabilityCVE-2026-61438updated 29 Oct 2026, 16:34 UTC

PraisonAI workflow runner can execute OS commands from YAML files

A malicious workflow file can run commands on the computer that runs it, as if the attacker were logged in there. Anyone who runs workflow files from untrusted sources is at risk.

Fixed in praisonai 4.6.78Official source
SeverityHighCVSS 7.3 of 10
Being exploited?UnclearNo reports either way yet
Fix available?Yespraisonai 4.6.78 · since 29 Oct 2026
How sure is this?Official sourceBacked by an official source

How the attack works simplified

  1. 1Attacker writes workflowYAML with script: import os
  2. 2File reaches runnerShared repo, template, or CI input
  3. 3PraisonAI runs itJob workflows enabled, AST check passes
  4. 4os.system runsShell commands run with caller's rights
  5. ResultAttacker runs commands on the host

What happened

The inline Python executor for YAML 'script:' steps in PraisonAI checks code with an AST allowlist and blocks some builtins, but it allows importing the 'os' module. Once imported, functions like os.system() and os.popen() are not blocked, so arbitrary shell commands run.

Why it is dangerous: An attacker who can supply a workflow file can run any command with the privileges of the process that executes it, which can mean full control of a CI runner, shared server, or developer machine.

Are you affected?

You are affected if you run PraisonAI versions before 4.6.78 with PRAISONAI_ALLOW_JOB_WORKFLOWS=true and execute YAML workflows (type: job) that may come from untrusted sources such as downloaded templates, shared repositories, or AI-generated workflows.

Package / productAffected versionsFixed in
praisonai · pip< 4.6.784.6.78

What to do

  1. Upgrade praisonai to version 4.6.78 or later.
  2. Keep PRAISONAI_ALLOW_JOB_WORKFLOWS unset or false unless you need job workflows.
  3. Only run workflow YAML files from trusted sources, and review any 'script:' steps before running them.

Timeline

  1. 29 Oct 2026, 16:34 UTC
    Publicly disclosed
  2. 29 Oct 2026, 16:34 UTC
    Reported by GitHub Advisory · GitHub Advisory
  3. 29 Oct 2026, 16:34 UTC
    Picked up by Zero Radar

Sources 1 report, oldest first

Official vendor, government CERT or advisory database

Technical details

ID
CVE-2026-61438 · GHSA-26mh-57q7-jfvr
CVSS
7.3 / 10
EPSS
0.3% chance of exploitation within 30 days
Disclosed
29 Oct 2026, 16:34 UTC
On Zero Radar
29 Oct 2026, 16:34 UTC (0 min after disclosure)
Last change
29 Oct 2026, 16:34 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

PraisonAI workflow runner can execute OS commands from YAML files · Zero Radar