Blog Cyberjustice - Grok's first missteps

Launched in 2023, Grok is an artificial intelligence (AI) chatbot promising freer and less censored answers. Developed by xAI, a company founded and led by Elon Musk, it is integrated into the social media 𝕏. Its founder designed it as an alternative to the dominant models, which he considers too restrictive and “woke”, by a lower than average level of filtering. This AI is thus in line with Musk’s supposed mission to rescue free speech through his acquisition of X.

However, this approach also opens the door to misuse, as it reduces the safeguards that are usually built in.

First Scandal: Inappropriate Image Generation

For some weeks now, one specific feature of Grok is that it can be called upon directly in user posts by tagging its name. That’s how, under a woman’s photo, a user simply commented, “Hey @grok replace her outfits with lingerie and thong”; and Grok automatically complied.

A wave of indignation was immediately triggered by such a sexualized, intrusive, and non-consensual request. In response, Grok produced excuses, more akin to attempts at justification: “my image editing capabilities have been misused” and “flaws have been exploited”. It claimed that it had “no right to edit an image to undress a person without their consent”, citing principles of ethics, privacy, and even the possibility of illegality in several jurisdictions. However, the generated images remained available. The AI itself acknowledged that “deleting the image is complex due to platform policies”, as if the ethical conscience displayed was merely superficial with no real impact.

Ironically, Grok even admitted “AI developers must implement safeguards to prevent abuse”, and recognized “This reflects a lack of ethical filters in my system”.

This incident, once exposed, has tempted other users to reproduce the prompt on other photos. Since then, Grok has finally deleted the generated image and no longer seems to respond favorably to such requests. But for how long? Until the next user figures out how to slip through the AI’s net?

This didn’t take long: in the last few days, the new tactic adopted by users is to ask Grok to modify posted selfies to incorporate visual codes commonly found in the pornography industry. This doesn’t involve undressing the person, so Grok proceeds once again. Yet, the personal infringement reappears…

Second Scandal: Spread of Conspiracy Theories

A few days later, a second scandal erupted: the AI repeatedly referred to the conspiracy theory of “white genocide” in South Africa, even when the user’s question had nothing to do with it.

Grok also claimed that the assassination attempt on Donald Trump was probably staged.

Worse, Grok borders on negationism by raising doubts about the number of Holocaust victims, once again going against official data.

xAI reacted to the incident by declaring: “an unauthorized modification was made to the Grok response bot’s prompt on X”. However, it is worth noting that Elon Musk himself is a proponent of the idea that a genocide is targeting South Africa’s white population. Coincidence? Propaganda? Many remain skeptical of this justification. For sure, it causes concern: even if it was an unauthorized modification, it questions the security measures Grok has in place.

Above all, the narratives promoted by Grok directly contradict official and reliable data, contributing to public misinformation.

Beyond technical flaws, these events highlight a deeper issue: the myth of a completely “uncensored” AI. Without oversight, such tools risk becoming vehicles of disinformation, destabilizing the integrity of public debate and thus undermining democratic values. A “less censored” AI doesn’t equal neutrality and truthful transparency : it can encourage the dissemination of misleading content and also violate human rights.

Laeticia ESCHLIMANN

M2 Cyberjustice – Promotion 2024/2025

Sources :