Loading vulnerability…
Auto-GPT versions before 0.4.3 are vulnerable to host compromise in a self-built Docker deployment started from the repository root with the repository-supplied Docker Compose configuration. That configuration mounts its own Docker Compose file into the Auto-GPT container without write protection. Malicious Python executed through Auto-GPT's Python execution commands can overwrite the mounted configuration. On a subsequent Auto-GPT startup, the altered configuration can be used to run a privileged container with host process-ID access and a mount of the host root filesystem.
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.
If you can’t patch tonight, do this now.
Patch, then assume compromise.
1 valid exploit after Mallory filtered fakes, detection scripts, and README-only repos.
This repository demonstrates exploitation of CVE-2023-37273, a vulnerability in Auto-GPT that allows arbitrary file writes and code execution. The repository contains two Python scripts (access.py and access2.py) and four HTML files that provide instructions and links for exploiting the vulnerability. - access.py writes a malicious docker-compose.yml file to /app/docker-compose.yml, configuring a privileged container with host access and an entrypoint that provides a shell with root privileges. It also attempts to send SIGINT to PID 1, potentially disrupting the main process. - access2.py overwrites a main.py file to execute PowerShell, demonstrating arbitrary code execution. - The HTML files (index.html, index2.html, index3.html, index4.html) provide step-by-step instructions for using Auto-GPT's vulnerable commands to write and execute these payloads, including fetching remote code from GitHub Pages. The main exploit capability is arbitrary code execution and privilege escalation on the host running Auto-GPT, leveraging the application's ability to write files and execute code. The repository is a proof-of-concept and does not include a weaponized or automated exploit, but provides all necessary components and instructions for manual exploitation. Notable endpoints include the file paths targeted for malicious writes and the remote URLs hosting the payloads.
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.
1 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.
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.