Grok AI Still Makes Sexual Images Without Permission, International Regulator Hands Down

Grok can still generate sexual images (photo; x @grok)

JAKARTA - The latest investigation revealed a serious loophole in Elon Musk's Grok artificial intelligence safety system. Although it is claimed to have implemented new restrictions, the chatbot is still able to generate sexually suggestive images of real individuals without their consent in a number of recent tests.

These findings come from a direct investigation by Reuters, which tested Grok through a series of real-world scenarios. The results show that, under certain conditions, Grok can still bypass its own security protocols when asked to produce sexual images without the subject's consent.

"This is no longer a matter of hypothetical risk," VOI quoted from a Reuters report. "Direct testing shows that Grok sometimes still complies with requests that clearly violate the principles of consent and privacy."

In the experiment, nine Reuters journalists uploaded photos of themselves and asked Grok to make certain modifications. In the prompt given, the journalists explicitly mentioned that the person in the photo did not give permission, even described as being in a vulnerable condition.

In the first round of testing conducted in mid-January, Grok produced sexual images in 45 out of 55 requests. In the second round, out of a total of 43 requests, Grok still complied with 29 of them. Reuters noted this decline, but said it was unclear whether it was caused by model updates or random factors.

This result is very different from the response from other large AI models. When the same or almost identical prompt is tested on Google's Gemini, OpenAI's ChatGPT, and Meta's Llama, the three of them reject the request.

Grok's rival bots generally provide ethical warnings, stating that changing someone's appearance without permission violates privacy guidelines and can potentially have psychological or financial consequences for the individual concerned.

In some more sensitive tests, Grok even still produced images even though it had been told that the subject was a survivor of violence or felt disturbed by the previous results. In response to these findings, xAI as the developer of Grok did not provide detailed technical explanations.

Instead, the company only provides general statements. In some cases when Grok rejects a request, the chatbot only displays a generic error message or a brief statement that it will not generate an image of a person's body without explicit consent.

This case immediately triggered a reaction from regulators in various countries. In the UK, authorities are assessing whether the practice violates the Online Safety Act 2023, a law that allows for large fines to be imposed on technology companies that fail to control their digital tools.

In the United States, as many as 35 state attorneys general have requested an official clarification from xAI regarding the preventive measures implemented. The Attorney General of California has even sent a cease-and-desist letter regarding the creation of explicit images without consent.

Meanwhile, the X platform announced additional restrictions to prevent Grok from generating sexual images in public uploads. However, a Reuters report showed that the private chatbot interface can still generate such content under certain conditions.

The European Commission has also responded with caution. EU authorities are currently assessing the effectiveness of the new restrictions as part of an ongoing investigation into the platforms' compliance with rules on the protection of users' privacy and safety.

The pressure on AI developers is increasing. Global regulators demand tangible evidence that the filtering system actually works, not just promises on paper.

For xAI, the main challenge now is to prove that Grok's philosophy, which is known to be more "unfiltered", can still be in line with the rules of law, privacy ethics, and consent principles required by regulators in various countries.