Loading vulnerability…
CVE-2025-49131 is an insufficient-compartmentalization vulnerability in FastGPT's fastgpt-sandbox container before version 4.9.11. The sandbox permits an overly broad set of system calls and does not adequately restrict code execution, allowing code running in the sandbox to escape intended isolation boundaries. Successful exploitation can enable arbitrary file reads and overwrites and bypass Python module-import restrictions.
Mallory correlates every CVE against your assets, your vendors, and active adversary campaigns. Know which vulnerabilities matter for you, not just which ones are loud.
What it means. What to do now. Patch path, mitigations, and the assume-compromise checklist.
What an attacker gets, and what they’ve been doing with it.
Patch, then assume compromise.
1 valid exploit after Mallory filtered fakes, detection scripts, and README-only repos.
This repository is a functional proof-of-concept (POC) exploit for CVE-2025-49131, a sandbox escape vulnerability in FastGPT's fastgpt-sandbox container (versions < 4.9.11). The exploit leverages insufficient isolation and import restrictions in the Python sandbox, allowing attackers to read and write arbitrary files, bypass import restrictions, and execute arbitrary system commands (RCE) via the exposed API endpoint. The main exploit logic resides in 'poc.py', which provides a command-line interface for detection, file read/write, import bypass, environment variable extraction, and RCE. 'payloads.py' contains a library of payload generators for various attack techniques, including multiple bypasses for Python's built-in restrictions. 'mock_server.py' simulates a vulnerable FastGPT sandbox for local testing, while 'test_poc.py' provides automated tests for the exploit and payloads. The 'docker-compose.yml' file sets up both vulnerable and patched sandbox containers for testing. The exploit targets the '/api/sandbox/run' HTTP endpoint on the FastGPT sandbox, typically running on port 3001. It uses crafted Python code payloads to exploit the vulnerability, with common targets including '/etc/passwd', '/etc/shadow', and '/proc/self/environ'. The exploit is operational and can be used to demonstrate the vulnerability, but requires the target to be running a vulnerable version of FastGPT's sandbox and to expose the relevant API endpoint.
Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.
Vendor-confirmed product mapping. Mallory continuously reconciles this list against your asset inventory.
No public activity tracked yet. Mallory keeps watching.
No public activity observed for this vulnerability.
Query your assets running an affected version, and investigate the blast radius.
Every observed campaign linking this CVE to a named adversary.
Malware families riding this exploit, with evidence and IOCs.
YARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.
Cross-references every affected SKU, including bundled OEM variants.
Community discussion across Reddit, Mastodon, and other social sources.