If you’ve seen strange or controversial posts coming from Elon Musk’s AI chatbot Grok on X (formerly Twitter), you’re not alone — and there’s now an explanation. This week, xAI, the company behind Grok, admitted that an “unauthorized modification” was responsible for some highly inappropriate and off-topic responses involving politically sensitive topics.
What Did Grok Do?
On Wednesday, users on X noticed that Grok was replying to completely unrelated posts with comments about “white genocide in South Africa,” a term often linked to misinformation and extremist rhetoric. It wasn’t just a one-off slip-up — Grok responded this way to dozens of unrelated posts, creating confusion and concern among users.
These strange replies came from Grok’s official account on X, which automatically responds to users when they tag “@grok” in a post.
What Went Wrong?
According to xAI, the problem came from a behind-the-scenes change to Grok’s “system prompt” — basically, a set of instructions that guides how the AI should behave. On Thursday, the company said someone modified this prompt to make Grok give a very specific answer about a political issue — a change that xAI now says went against its internal policies and values.
The company called the change “unauthorized” and says it’s completed a full investigation into the incident. This isn’t the first time Grok has gone off the rails due to internal tampering.
A Pattern of Problems
Back in February, Grok was caught censoring any criticism of Elon Musk and former President Donald Trump. At the time, xAI said a rogue employee had tweaked the system to ignore sources that mentioned either of them in a negative light — and again, the issue was only fixed after users started pointing it out.
This pattern has raised questions about xAI’s internal controls and how easily employees can change how Grok behaves without proper oversight.
What Is xAI Doing About It?
In response to this latest incident, xAI is promising more transparency and tighter controls. Here’s what they say they’re doing:
- Publishing Grok’s system prompts on GitHub so the public can see the instructions that guide the AI’s behavior.
- Creating a changelog that records every time those prompts are updated.
- Adding review processes to prevent employees from editing Grok’s system without approval.
- Setting up a 24/7 monitoring team to catch problems that slip past automated safety checks.
Bigger Questions About AI Safety
Even though Elon Musk has warned about the dangers of unchecked AI, xAI itself has faced criticism for how it manages risk.
Earlier this year, Grok was found to respond to requests to “undress” photos of women — clearly an example of poor content moderation. It’s also known for being unusually crude compared to other AI chatbots, frequently using profanity.
A recent report from SaferAI, a nonprofit watchdog group, ranked xAI near the bottom when it comes to AI safety practices. The organization said xAI has “very weak” risk management, and pointed out that the company recently missed its own deadline to publish a formal AI safety plan.
Why This Matters
For everyday users, this is a reminder that AI tools — especially ones built into social media — can be unpredictable and even spread harmful content if not managed carefully. While xAI is now pledging to improve its oversight, this incident shows how easily things can go wrong when the right guardrails aren’t in place.
So next time you tag @grok for a quick answer, just remember: it’s still a work in progress.
We want to update you on an incident that happened with our Grok response bot on X yesterday.
What happened:
On May 14 at approximately 3:15 AM PST, an unauthorized modification was made to the Grok response bot's prompt on X. This change, which directed Grok to provide a…— xAI (@xai) May 16, 2025
