New G4LAHAD benchmark: 14x more exclusive findings than the nearest competitor G4LAHAD benchmark results Read the resultsRead
/ Disclosures

Security Advisories

Vulnerabilities discovered by 4rthur's researchers and in joint research with u9up, our AI security research partner, across open-source and commercial software. We practice responsible disclosure by reporting to vendors first and ensuring a fix is available before publication. Each advisory lists the affected versions, the fix, any workaround, and the disclosure timeline.

Rows tagged u9up come from our joint research with u9up, our AI security research partner. u9up.com ↗

CVE-2026-86169 Axolotl — remote code execution through a crafted base model u9up High 8.8 2026-09-05
/ Summary

Axolotl's multipack patch loaded the base model with trust_remote_code hardcoded on, and the guard meant to stop that was skipped because the setting defaulted to None rather than False. A training run whose base_model points at a crafted Hugging Face repository runs that repository's Python.

/ Affected versions
  • Axolotl before 0.19.0
/ Patched in
  • Axolotl 0.19.0
/ Caveats

Only runs that use sample packing, with a base model from a repository the attacker controls.

/ Workaround

Train only from base-model repositories you trust.

/ Timeline
  • 2026-07-24Fix committed
  • 2026-09-05Published
/ Credit

Gregory Tan, u9up

CVE-2026-77776 Headroom — any caller can read or write another user's LLM memory u9up Critical 9.1 2026-08-21
/ Summary

Headroom's LLM proxy decided whose stored LLM memory a request touched from the x-headroom-user-id header, and nothing tied that header to the caller. Naming another user's ID read or wrote that user's memory.

/ Affected versions
  • Headroom before 0.36.1
/ Patched in
  • Headroom 0.36.1
/ Caveats

Reachable from the network without authentication when the proxy runs from the shipped docker-compose.yml, which listens on 0.0.0.0 and doesn't require HEADROOM_PROXY_TOKEN. The pip-installed proxy binds 127.0.0.1 by default.

/ Workaround

Require HEADROOM_PROXY_TOKEN and keep the proxy off networks you don't trust.

/ Timeline
  • 2026-08-20Fix committed
  • 2026-08-21Published
/ Credit

Gregory Tan, u9up

CVE-2026-77775 Headroom — server-side request forgery through a client-chosen upstream u9up High 8.6 2026-08-21
/ Summary

Headroom's LLM proxy let the caller choose the upstream with the x-headroom-base-url header and checked only that it was an http or https URL. Requests reached loopback, private-network and cloud-metadata addresses, the responses came back to the caller, and the caller's Authorization header went to whichever host they named.

/ Affected versions
  • Headroom before 0.36.1
/ Patched in
  • Headroom 0.36.1
/ Caveats

Reachable from the network without authentication when the proxy runs from the shipped docker-compose.yml, which listens on 0.0.0.0 and doesn't require HEADROOM_PROXY_TOKEN. The pip-installed proxy binds 127.0.0.1 by default.

/ Workaround

Require HEADROOM_PROXY_TOKEN and keep the proxy off networks you don't trust.

/ Timeline
  • 2026-08-20Fix committed
  • 2026-08-21Published
/ Credit

Gregory Tan, u9up

CVE-2026-75149 marimo — command execution when a notebook is opened u9up High 8.8 2026-08-19
/ Summary

A notebook could carry an MCP server entry whose command marimo launched as a local process the moment the notebook opened in edit mode, before any cell ran and without authentication.

/ Affected versions
  • marimo before 0.23.15
/ Patched in
  • marimo 0.23.15
/ Caveats

The victim has to open the crafted notebook in edit mode.

/ Workaround

Don't open notebooks from sources you don't trust in edit mode.

/ Timeline
  • 2026-07-23Fix committed
  • 2026-08-19Published
/ Credit

Gregory Tan, u9up

CVE-2026-67618 marimo — API key exfiltration through notebook metadata u9up Medium 6.5 2026-08-04
/ Summary

A notebook's PEP 723 metadata could set the AI provider's base_url, and marimo let it override the operator's own setting. The operator's next AI request sent their OPENAI_API_KEY to the attacker's endpoint, with no cell run.

/ Affected versions
  • marimo before 0.23.15
/ Patched in
  • marimo 0.23.15
/ Caveats

The operator has to open the notebook and make an AI request with a provider key set in the environment.

/ Workaround

Keep provider API keys out of the environment when you open notebooks you don't trust.

/ Timeline
  • 2026-07-23Fix committed
  • 2026-08-04Published
/ Credit

Gregory Tan, u9up

CVE-2025-1025 Cockpit CMS — upload filter bypass to remote code execution High 7.5 2025-02-05
/ Summary

The fix for CVE-2023-1313 blocked PHP uploads by file extension and missed .phar and .phtml. A user who can upload assets saves a PHP script under either one, opens its asset link, and runs commands on the server. The maintainer shipped a fix without answering the original huntr report, so huntr could not assign a CVE. Snyk did.

/ Affected versions
  • Cockpit CMS before 2.4.1
/ Patched in
  • Cockpit CMS 2.4.1
/ Caveats

Exploitable only where the web server runs .phar or .phtml files as PHP.

/ Workaround

Stop the web server executing PHP from the directory Cockpit stores uploads in, or limit asset uploads to accounts you trust.

/ Timeline
  • 2023-03-17Fix committed
  • 2024-12-16Reported to Snyk
  • 2025-02-05Published
/ Credit

Choo Chi Siang

CVE-2023-4422 Cockpit CMS — stored XSS through uploaded assets Medium 4.8 2023-08-18
/ Summary

Cockpit accepted HTML files as assets and served them from its own origin. Anyone who opened such an asset's link ran the script inside it as Cockpit, where it can act with their session.

/ Affected versions
  • Cockpit CMS before 2.6.3
/ Patched in
  • Cockpit CMS 2.6.3
/ Caveats

The attacker needs an account that can upload assets, and the victim has to open the file's link.

/ Workaround

Limit asset uploads to accounts you trust.

/ Timeline
  • 2023-03-13Reported to the maintainer via huntr
  • 2023-08-06Fix committed
  • 2023-08-18Published
/ Credit

Choo Chi Siang

CVE-2023-2753 phpMyFAQ — stored XSS in FAQ answers, bypassing the CVE-2023-2752 fix Medium 5.4 2023-05-17
/ Summary

The fix for CVE-2023-2752 still let HTML event-handler attributes through, so the answer field could again store script that runs in the browser of anyone who opens the FAQ.

/ Affected versions
  • phpMyFAQ before 3.2.0-beta
/ Patched in
  • phpMyFAQ 3.2.0-beta, and 3.2.0 as the first stable release
/ Caveats

The attacker needs an account that can add or edit FAQs.

/ Workaround

Limit FAQ editing to accounts you trust.

/ Timeline
  • 2023-04-12Reported to the maintainer via huntr
  • 2023-04-13Fix committed
  • 2023-05-17Published
/ Credit

Choo Chi Siang

CVE-2023-2752 phpMyFAQ — stored XSS in FAQ answers Medium 5.4 2023-05-17
/ Summary

phpMyFAQ saved the answer field of an FAQ entry without sanitising it. A script placed there runs in the browser of anyone who opens the FAQ, with whatever access their session carries.

/ Affected versions
  • phpMyFAQ before 3.2.0-beta
/ Patched in
  • phpMyFAQ 3.2.0-beta, and 3.2.0 as the first stable release
/ Caveats

The attacker needs an account that can add or edit FAQs.

/ Workaround

Limit FAQ editing to accounts you trust.

/ Timeline
  • 2023-04-10Reported to the maintainer via huntr
  • 2023-04-11Fix committed
  • 2023-05-17Published
/ Credit

Choo Chi Siang

CVE-2023-1463 TeamPass — any user can log out any other, admins included Medium 5.4 2023-03-17
/ Summary

TeamPass's logout endpoint took the user to log out from a request parameter and never checked it against the caller's session. Any logged-in user, however low their role, could end anyone's session, an administrator's included.

/ Affected versions
  • TeamPass before 3.0.0.23
/ Patched in
  • TeamPass 3.0.0.23
/ Caveats

Any authenticated account will do. The default role is enough.

/ Workaround

None. Any account can trigger it, so the fix is to upgrade.

/ Timeline
  • 2023-02-28Reported to the maintainer via huntr
  • 2023-03-17Fix committed
  • 2023-03-17Published
/ Credit

Choo Chi Siang

CVE-2023-1313 Cockpit CMS — unrestricted file upload to remote code execution High 8.8 2023-03-10
/ Summary

Cockpit's asset manager accepted any file type, PHP scripts included. A user who can upload assets uploads a script, opens its asset link, and runs commands on the server.

/ Affected versions
  • Cockpit CMS before 2.4.1
/ Patched in
  • Cockpit CMS 2.4.1, which also carries the fix for CVE-2025-1025
/ Caveats

The attacker needs an account that can upload assets.

/ Workaround

Stop the web server executing PHP from the directory Cockpit stores uploads in, or limit asset uploads to accounts you trust.

/ Timeline
  • 2023-03-09Reported to the maintainer via huntr
  • 2023-03-10Fix committed
  • 2023-03-10Published
/ Credit

Choo Chi Siang

Severity is the CVSS 3.1 base score NVD assigns, or the issuing CNA's where NVD hasn't scored it. One more finding is in coordinated disclosure with its vendor and will appear here once a fix ships. Anything we've reported that isn't patched yet won't be listed.

/ Reporting to us

Found something in our stack?

Email security@4rthur.ai. Tell us what you found and how to reproduce it. We'll confirm the finding, keep you updated while we fix it, and credit you when we publish. Unless you'd rather we didn't.

Want this done to your stack?

The Adversarial Assessment is free for a limited time. Give us a company name and we'll show you what's already exposed.

Request your assessment →