- A security researcher discovered a disturbing API key leakage
- The flight would come from Doge Marko Elez staff
- This is not the first security problem from Doge
A staff member having access to the personal data of millions of Americans apparently disclosed the key to the API to at least four dozen LLMs developed by the XAI artificial intelligence company, including the Xa Grok chatbot (formerly Twitter).
Security expert Brian Krebs has revealed that Marko Elez, an employee of the Elon Musk government efficiency department, had access to databases sensitive to the US Social Security Administration, justice and cash services in the context of Doge’s work to “ rationalize ” the ministries to increase efficiency.
Ironically, researchers have recently discovered that the references of a DOGE worker have been exposed by infostative malware, so the DOGE security file is so far less than impressive.
Grok exposed
A code script was attached to Github named ‘Agent.Py’ which included an application programming interface key (API) for XAI by Elez. This was first reported by Gitguardian, a company that scans GitHub for API secret tokens, information database identification and certificates – and alerts affected users.
The API key exposed made it possible to access at least 52 LLM different used by XAI, the most recent being an LLM called “Grok 4-0709”, created on July 9, 2025 – according to the director of safety of security of Seralys, Philippe CatureGli.
CatureGli warned KrebssonSecurity: “If a developer cannot keep a private API key, this raises questions about how they manage government information much more sensitive to closed doors.”
The code repository which contains the private API key has since been deleted after Elez has been informed by e-mail of the leak, however, the key still works and has not yet been revoked, so the problem is far from resolved.
This is not the first time that internal XAI APIs have been disclosed, with LLM designed for other Musk organizations, such as SpaceX, Tesla and Twitter / X exhibited earlier in 2025, Krebs confirmed.
“A leak is an error,” said CatureGli, “but when the same type of sensitive key is exposed again and again, it is not only bad luck, it is a deeper sign of negligence and a broken security culture.”
