CVE-2026-34938
Sandbox Bypass in PraisonAI Agents Enables Remote Code Execution
Publication date: 2026-04-03
Last updated on: 2026-04-14
Assigner: GitHub, Inc.
Description
Description
CVSS Scores
EPSS Scores
| Probability: | |
| Percentile: |
Meta Information
Affected Vendors & Products
| Vendor | Product | Version / Range |
|---|---|---|
| praison | praisonaiagents | to 1.5.90 (exc) |
Helpful Resources
Exploitability
| CWE ID | Description |
|---|---|
| CWE-693 | The product does not use or incorrectly uses a protection mechanism that provides sufficient defense against directed attacks against the product. |
Attack-Flow Graph
AI Powered Q&A
How can this vulnerability impact me? :
This vulnerability allows an attacker to achieve arbitrary OS command execution on the host running PraisonAI. This means an attacker can run any command on the affected system with potentially full control, leading to complete compromise of the host, data theft, data destruction, or further attacks within the network.
What immediate steps should I take to mitigate this vulnerability?
The vulnerability has been patched in PraisonAI version 1.5.90. Immediate mitigation involves upgrading PraisonAI to version 1.5.90 or later.
Can you explain this vulnerability to me?
PraisonAI is a multi-agent teams system. Before version 1.5.90, the execute_code() function in praisonai-agents runs attacker-controlled Python code inside a three-layer sandbox. However, this sandbox can be completely bypassed by using a string subclass with an overridden startswith() method passed to the _safe_getattr wrapper. This bypass allows an attacker to execute arbitrary operating system commands on the host machine.