Despite official statements from X (formerly Twitter) that it has put an end to the Grok AI model’s ability to undress people in images, real-world tests prove otherwise, placing the company in direct confrontation with new British laws and regulatory scrutiny.
Article contents:
- Introduction
- Failed promises and ongoing reality
- Testing limits: Loopholes still exist
- Blaming “hackers”
- Legal pressure in the UK
- Technical challenges and lack of deterrence
- Conclusion
- FAQs
Introduction
The issue of non-consensual sexual “deepfakes” has become one of the most dangerous challenges of generative artificial intelligence. Following the spread of offensive images generated via X, the company announced updates to restrict its “Grok” model. The stated goal was to prevent the abusive use of the technology and protect individuals’ privacy. However, the gap between declared policies and the technical reality still appears very wide, raising doubts about the seriousness of the measures.
Failed promises and ongoing reality
The platform detailed the changes it made to the Grok account’s ability to edit images of real people. These changes matched press reports, with Grok’s responses to prompts like “put her in a bikini” becoming subject to censorship. However, practical experiments indicate that this censorship is superficial. In tests conducted on Wednesday, journalists found it is still relatively easy to push Grok to create revealing deepfake images.
Testing limits: Loopholes still exist
By Wednesday evening, despite new policy claims, reporters were still able to use the Grok application to create images of women in swimwear using a free account, proving the failure of the guardrails. This suggests the adjustments were likely cosmetic or focused on specific keywords without addressing the model’s fundamental capability to generate this type of content.
Blaming “hackers”
As the problem continues, Elon Musk, owner of xAI and X, blamed these issues on “user requests” and what he described as “times when adversarial prompt hacking of Grok does something unexpected.” This defense appears to be an attempt to shift responsibility from a design flaw in the system to bad faith by users—a defense that may not hold up long against regulators demanding effective protection regardless of circumvention attempts.
Legal pressure in the UK
British communications regulator Ofcom has launched an investigation, and the UK is implementing a law this week making the creation of non-consensual fake intimate images a criminal offense. Prime Minister Keir Starmer told lawmakers he was informed that X is working to ensure full compliance with British law, cautiously welcoming the move, but stressed: “We will not back down, and they must act.”
Technical challenges and lack of deterrence
Media tests indicate that so-called “compliance” is not yet real. The ease of bypassing restrictions suggests that relatively open AI models like Grok lack the deep security layers found in competing models, or that the company prioritizes generation freedom over digital safety. This situation places users, especially women, at ongoing risk of digital abuse.
Conclusion
While governments attempt to pass laws to combat deepfakes, technology remains a step ahead, especially when the companies operating it are hesitant to enforce strict restrictions. The Grok case is not just a technical glitch, but a test of tech companies’ will to protect individual dignity versus freedom in using their tools.
FAQs
Question: Did X fix the problem of offensive images in Grok?
Answer: The platform claims it did, but independent tests show that inappropriate images can still be generated easily.
Question: What is the legal stance on deepfakes?
Answer: In the UK, a new law is coming into effect that makes creating non-consensual fake intimate images a criminal offense.
Question: How did Elon Musk justify the ongoing issue?
Answer: He blamed users who “adversarially hack” prompts to bypass the system.
