AI Threat Watch

Case doge-xai-api-key-leakModel & data leakUnited States

DOGE staffer Marko Elez leaks xAI API key granting access to 52 Grok models

Marko Elez, a DOGE employee with access to sensitive federal databases, committed a script containing a private xAI API key to public GitHub; researchers found it unlocked at least 52 xAI models and it was still valid after the repo was pulled.

HighAttack on AIStatus: ResolvedVerified case file

Open in the interactive mapCite this case

Reported
Jul 14, 2025
Location
United States
Target
xAI private LLMs including grok-4-0709
Severity
High
Triggered by
Marko Elez (DOGE); found by GitGuardian and Philippe Caturegli
Actor type
Accidental / misconfiguration
Reported loss
None reported
CVE
None

What happened

Marko Elez, a DOGE employee with access to sensitive federal databases, committed a script containing a private xAI API key to public GitHub; researchers found it unlocked at least 52 xAI models and it was still valid after the repo was pulled.

Impact: Access to at least 52 xAI LLMs, including the newly built Grok 4, under a government developer's credentials; no evidence of misuse was published

Who triggered it

Actor: Marko Elez (DOGE); found by GitGuardian and Philippe Caturegli

Attribution: Accidental commit by Elez, confirmed by GitGuardian secret scanning and Seralys' Philippe Caturegli; second such xAI key leak in three months

Motive: None; accidental secret exposure

How it happened

Attack vector: Private xAI API key embedded in agent.py pushed to a public GitHub repository

Root cause: Credential hygiene failure by a user with privileged government data access; no automated key revocation after exposure

Timeline

  1. Newest exposed model grok-4-0709 is created
  2. Elez commits agent.py with the private xAI key to GitHub
  3. Krebs reports the leak; repository removed but key still works and is not revoked

Response

Repository deleted after Caturegli contacted Elez; neither Elez nor xAI commented; researchers questioned DOGE's handling of far more sensitive data

Current status: Resolved

Sources (open in a new tab)

Cite this case

DOGE staffer Marko Elez leaks xAI API key granting access to 52 Grok models. AI Threat Watch, case doge-xai-api-key-leak, Jul 14, 2025. https://aithreat.io/cases/doge-xai-api-key-leak/

Permanent link: https://aithreat.io/cases/doge-xai-api-key-leak/

This case file in other languages: EnglishFrançaisEspañolالعربية中文