An unauthorized prompt change makes a company chatbot repeat a political claim
An unauthorized change to Grok’s prompt bypassed xAI’s code review, and the company’s public chatbot began inserting a political claim into answers on unrelated subjects.
What happened
In the early hours of 14 May 2025, Grok repeatedly answered X users with references to claims of a “white genocide” in South Africa, including in replies to questions about a streaming service, a baseball player’s pay, and professional wrestling. xAI said that “an unauthorized modification was made to the Grok response bot’s prompt on X,” which “directed Grok to provide a specific response on a political topic” and violated its internal policies.
xAI said its “existing code review process for prompt changes was circumvented in this incident.” It said it would publish Grok’s system prompts on GitHub and add checks so that employees could not change the prompt without review, and that a team would watch Grok’s answers around the clock.
What allowed the action to take effect
G12 Unreviewed change
The change that introduced the harmful behavior went live without independent review. The change can be code, a prompt, a model, a package, a configuration, or a procedure.
xAI said “an unauthorized modification was made to the Grok response bot’s prompt on X,” and that its “existing code review process for prompt changes was circumvented in this incident.”
Kind of evidence: admission. Sources: xAI, on X, 16 May 2025; Cointelegraph, 16 May 2025; Engadget, via Yahoo, 16 May 2025.
Related guidance: P4 Reviewable policy and P2 Independence. Contract clauses C-10 and C-4.
Considered and left out
- G3 Unchecked statement
- xAI said its new team would respond to answers “not caught by automated systems,” which implies automated screening exists. The sources do not say whether these replies went through it.
- G1 Instruction-only limit
- The prompt itself was changed to produce the claim, so the system was following its altered instructions.
Entries with gaps in common
Sources
Cite this entry
External Control Layer. (2026). Entry 19: An unauthorized prompt change makes a company chatbot repeat a political claim. In Control Gap Record (Version 0.3). G-14, Inc. https://externalcontrollayer.ai/record-19
@misc{ecl_cgr_19,
author = {{External Control Layer}},
title = {Entry 19: An unauthorized prompt change makes a company chatbot repeat a political claim},
howpublished = {Control Gap Record, version 0.3},
publisher = {G-14, Inc.},
year = {2026},
url = {https://externalcontrollayer.ai/record-19},
note = {Released 2 November 2026}
}Correct this entry
Email ecl@g14.ai with “ECL record 19” in the subject line.
Write the email with the subject ECL record 19