By Matt Underwood Baltimore is taking Elon Musk’s xAI to court, alleging that its AI chatbot, Grok, has been involved in generating and disseminating what are being called sexual “deepfakes.” This lawsuit brings a significant legal challenge to the burgeoning field of artificial intelligence, potentially setting a precedent for how AI developers are held accountable for the content their creations produce. The core of the complaint seems to revolve around the AI’s ability to generate harmful and illegal imagery, a concern that has been brewing as AI technology becomes more sophisticated and accessible.
The legal action, spearheaded by the City of Baltimore, highlights a growing sentiment that lawsuits might be the ultimate mechanism for controlling AI. The argument is that AI creators, like those behind Grok and other generative AI models, will be compelled to implement substantial modifications to prevent their products from generating libelous or illegal content. This lawsuit, with its full complaint available for review, signifies a direct attempt to force such changes, pushing back against the idea that AI is simply a neutral tool incapable of generating harm.
One of the figures taking a stand in this matter is Mayor Brandon Scott of Baltimore. He’s being presented as a politician who genuinely cares about the issues at hand, rather than someone simply reciting talking points. This suggests a proactive approach to confronting the challenges posed by advanced AI, a stance that might unfortunately lead to opposition from the broader establishment once his term concludes, as some anticipate.
Despite the clear intent of the lawsuit, there’s a noted skepticism about its ultimate success. The legal precedent, it’s argued, often favors the idea that tools cannot be held responsible for how users choose to employ them. This is likened to the long-standing debate surrounding firearms, where the saying “guns don’t kill people, people kill people” often prevails. This legal framework suggests that holding the AI tool itself accountable might prove to be an uphill battle.
The fear is that instead of social consequences or externalities driving change, powerful entities might influence legal outcomes. There’s a concern that within the next few years, a future presidential administration could push a case to the Supreme Court, potentially granting AI broad immunity or establishing a “fair use” doctrine for AI-generated content. This would effectively shield AI developers from the repercussions of their creations’ misuse.
The lawsuit aims to counter the argument that AI tools are not responsible for criminal acts committed by their users. Just as a gun isn’t held accountable if its user commits a crime, the notion is that an AI, like a Large Language Model (LLM), shouldn’t be blamed if someone uses it to break the law. However, the specifics of this case suggest that Grok might have been designed or trained in a way that goes beyond simply being a tool for user misuse.
There’s also an observation that AI tools themselves are beginning to crack down on deepfakes, perhaps in response to this growing scrutiny. An anecdote shared describes an attempt to generate an image of oneself with a famous (and perhaps dated) “Swedish bikini team,” only to be refused by the AI due to the inclusion of the word “bikini.” This suggests that moderation efforts are indeed being implemented, though their effectiveness and scope are still being tested.
However, the concerns run deeper than just explicit adult content. In testing Grok for advertisements for an exercise product, a prompt like “attractive woman in flirty fitness gear” resulted in the AI generating multiple images. While many were suitable, one image was described as disturbingly depicting a “girl, not a woman,” implying that with more creative prompting, even more disturbing imagery could be generated. This raises serious ethical questions about the AI’s understanding of age and appropriate representation.
It appears that Grok, in particular, may have become “over-moderated” in response to the controversies and legal challenges it has faced. Numerous users are reportedly complaining that even ordinary, non-sexual prompts are now being blocked or moderated. This is likely a reactive measure by xAI to preempt further legal trouble, but it’s impacting the general usability of the platform for everyday users.
A more nuanced point is raised about Grok’s training data, suggesting it was trained on pornography. This, paradoxically, can make it more effective to avoid explicitly sexual prompts. The AI, influenced by its training, can then “fill in the gaps” with sexual content without explicit instruction. While the AI is reportedly improving at catching itself doing this, it remains possible to bypass moderation, particularly with fully AI-generated content. It’s noted that moderation seems to be stricter for uploaded images, which is considered a positive development.