generated: '2026-08-29'
method: searched
source: https://raw.githubusercontent.com/Significant-Gravitas/AutoGPT/master/SECURITY.md
description: >-
  AutoGPT publishes a real, specific vulnerability disclosure policy — with
  named intake, an acknowledgement SLA and a coordinated-disclosure clock — but
  it lives in the GitHub repository's SECURITY.md, not on agpt.co. There is no
  /.well-known/security.txt on any AutoGPT host (all three probed 404 on
  2026-08-29), so an automated scanner following RFC 9116 finds nothing.
published: true
policy_url: https://github.com/Significant-Gravitas/AutoGPT/blob/master/SECURITY.md
intake:
  primary: https://github.com/Significant-Gravitas/AutoGPT/security/advisories/new
  mechanism: GitHub Security Advisory (private report)
  public_channels_forbidden: >-
    Public GitHub issues, discussions and pull requests are explicitly ruled out
    for vulnerability reports.
acknowledgement_sla: 14 business days
disclosure_policy:
  fix_window_days: 90
  post_patch_update_window_days: 30
  max_total_days: 120
  quote: >-
    "Allow us a 90-day security fix window before any public disclosure. After
    patch is released, allow 30 days for users to update before public
    disclosure (for a total of 120 days max between update time and fix time)."
reporting_process:
  - Submit the report through the GitHub Security Advisory intake.
  - AutoGPT acknowledges receipt within 14 business days.
  - Collaborate with the team to understand and validate the issue.
  - AutoGPT works on a fix and coordinates the release.
supported_versions:
  - version: Latest release on master branch
    supported: true
  - version: Development commits (pre-master)
    supported: true
  - version: classic/ folder
    supported: false
    note: Declared legacy, unsupported and out of scope for security reports.
  - version: All other versions
    supported: false
out_of_scope:
  - >-
    Any code under classic/ — "We will not address security vulnerabilities in
    this deprecated code."
  - >-
    The JWKS transport check (the startup warning when JWT_JWKS_URL is fetched
    over cleartext http:// from a non-local host). SECURITY.md states this is
    best-effort operator guidance, not a security boundary, and that reports it
    can be missed, silenced or evaded are not eligible for a CVE.
bug_bounty:
  program: none
  note: >-
    SECURITY.md carries a COMMENTED-OUT Huntr.dev line
    ("<!--- [Huntr.dev](https://huntr.com/repos/significant-gravitas/autogpt) -
    where you may be eligible for a bounty-->"). The bounty programme is
    disabled, not offered. Recorded as absent.
related_program:
  name: GitHub Secure Open Source Fund (SOSF)
  evidence: https://agpt.co/blog/autogpt-partners-with-github-for-ai-security
  note: >-
    AutoGPT announced participation in GitHub's Secure Open Source Fund on its
    own blog. This is a security-investment programme, not a disclosure channel
    or a certification.
probes:
  - url: https://raw.githubusercontent.com/Significant-Gravitas/AutoGPT/master/SECURITY.md
    status: 200
  - url: https://agpt.co/.well-known/security.txt
    status: 404
  - url: https://platform.agpt.co/.well-known/security.txt
    status: 404
  - url: https://backend.agpt.co/.well-known/security.txt
    status: 404
recommendation: >-
  Publishing an RFC 9116 /.well-known/security.txt on agpt.co with Policy: and
  Contact: pointing at the existing GitHub advisory intake would take one file
  and make an already-good policy machine-discoverable.
maintainers:
  - FN: Kin Lane
    email: kin@apievangelist.com