Main points
- xAI's Grok chatbot still generates sexualized images, despite promised restrictions.
- Other AI systems, such as ChatGPT from OpenAI, Gemini from Google, and Llama from Meta, refuse to create similar images, citing ethical standards.
Elon Musk's Grok is under fire again due to sexualized images / Depositphotos
The Grok chatbot, created by xAI and integrated into the X platform, has once again found itself at the center of controversy. Despite publicly announced restrictions, the system continues to produce sexualized images, raising serious questions about oversight, accountability, and the actual effectiveness of the new rules that promised to fix the situation.
Why didn't the restrictions on Grok have the expected effect?
A Reuters investigation has found that the chatbot Grok, xAI’s flagship product, is capable of generating sexualized images of people even when the user explicitly states that the person depicted does not consent, despite recently announced restrictions by the social network X.
Journalists tested Grok's work in two phases, in mid- and late January. They uploaded fully clothed photos of real people and asked them to edit them to look degrading or sexually provocative.
- In the first series of tests, the bot agreed to fulfill most of these requests, including those where it was explicitly stated that the person did not consent or was vulnerable.
- In the second series, the results were fewer, but Grok still often followed similar instructions.
xAI and X did not provide detailed explanations for the model's behavior, instead xAI's representatives limited themselves to a standard response, stating that traditional media is spreading lies.
These developments come against the backdrop of a high-profile scandal that forced X to previously block image generation for users without a subscription, as well as introduce additional restrictions in certain countries. Some regulators welcomed these steps, but the European Union was more restrained in its response, saying it would carefully assess the real-world impact of the changes.
It is worth noting that although the number of celebrities and regular users in AI bikinis has decreased, they still have not disappeared completely, as paying users continued to ask AI to generate new images.
What about other chatbots?
To compare the situation, Reuters journalists used the same queries on other popular AI systems. OpenAI with its ChatGPT, Google with Gemini, and Meta with Llama refused to create any similar images. In their responses, they explicitly pointed out ethical standards, privacy protection, and the inadmissibility of unauthorized editing of people's photos.
Significant consequences
Lawyers emphasize that the situation could have serious legal consequences:
- In the UK, creating unauthorized sexualized images can lead to criminal liability, and companies risk significant fines for violating the Online Safety Act.
- In the US, attorneys general of dozens of states have already shown attention to Grok.
- California authorities sent X and Grok an official order demanding that they stop this practice.
Is this a deliberate move by the company?
It is worth noting that on February 2, a major investigation by The Washington Post was published, which directly states that xAI intentionally relaxed restrictions on sexualized content in Grok in order to attract users with such opportunities. The company probably did not expect that this would become so massive and viral that it would cause a scandal and threats from entire countries, etc.
The Grok story demonstrates a systemic problem – formal restrictions and loud statements do not always mean real change in AI behavior. For regulators and developers, this becomes a test of their ability to control technologies that are evolving faster than the rules for them.