xAI’s Grok-2, Elon Musk’s latest answer to ChatGPT, is making waves and sparking debate. On its website, xAI claims Grok-2 performs better than OpenAI’s GPT-4 and Anthropic’s Claude. However, early adopters have found that the A.I. model’s lack of content moderation and safety guardrails has created harmful, misleading and sometimes legally questionable content.
Currently, Grok-2 is only available in beta for X Premium users. As users began testing the tool’s boundaries, they discovered that Grok-2 could generate images that rival tools such as Midjourney or Google Gemini often prohibit, from deepfakes of public figures to violent and sexually explicit content. Examples include an image of Donald Trump hugging a pregnant Kamala Harris and one depicting Mickey Mouse and Pikachu engaging in inappropriate behavior.
Legal experts warn that Grok-2’s advanced capability and lack of safety guardrails could pose serious risks to its users and society. “Grok-2’s ability to create deepfakes that are almost indistinguishable from authentic could have serious negative effects on areas like personal privacy, criminal justice, anti-discrimination laws and data privacy regulations,” Randy McCarthy, an intellectual property attorney, told Observer.
Elon Musk’s A.I. dilemma
Although Elon Musk has been vocal about his concerns with unregulated A.I., Grok-2 might just be falling into the pitfalls he once warned against. Recently, Musk tweeted his support for California’s Frontier Artificial Intelligence Models Act (SB 1047), the first significant legal framework in the U.S. to regulate A.I. The bill is currently awaiting Governor Gavin Newsom’s signature to become a state law, a move that Musk considers crucial for promoting ethical A.I. development. This is in stark contrast to Grok-2, however, which appears to lack the level of safety guardrails seen in rival apps.
“On the one hand, Elon preaches the need for A.I. regulation and is among the most vocal superintelligence doomsayers. On the other, [xAI] released a relatively unrefined A.I. model. Clearly, he values uncensored free speech,” Brandon Purcell, vice president and a principal analyst at Forrester specializing in A.I., told Observer. “Companies could learn from Elon on this one, as many haven’t clearly defined their values yet. If you don’t spell out your values, A.I. will end up doing it for you—and you probably won’t like how it turns out.”
Many tech companies are currently developing methods to label A.I.-generated content. For instance, Adobe and Microsoft use a symbol from the Coalition for Content Provenance and Authenticity (C2PA), a global standards body certifying media content’s source and history, to mark A.I.-generated images. Musk’s dismissal of these concerns with tweets like “Grok is the most fun A.I. in the world!” has only added fuel to the fire, signaling a lack of urgency in addressing the risks of the technology.
Fixing Grok-2 would require a fundamental shift in how xAI approaches A.I. development, legal and tech industry experts say. “One possible solution is for A.I. systems to include embedded metadata or some form of encoding in the outputs to clearly identify them as A.I.-generated,” McCarthy said.
In addition, “intellectual property rules will need to evolve to handle the unique challenges that A.I. brings to the table,” he added. “At the same time, we should think about new revenue models—like ways to fairly compensate artists whose work might be used to train these A.I. systems and figuring out how to navigate this changing landscape responsibly.”
“To prevent the misuse of A.I. tools like Grok-2, businesses should be taking a proactive approach by implementing comprehensive usage policies that clearly outline acceptable use cases for their technologies,” Max Li, CEO of OORT, a decentralized cloud computing platform, told Observer. “In addition, continuous and vigilant monitoring of the outputs generated by A.I. systems is crucial to identify and address harmful uses of the technology before they escalate.”
