Elon Musk's Grok continues to do humanity a solid by (accidentally) illustrating why AI needs meaningful guardrails. The xAI bot's latest demonstration is detailed in a pair of reports by Futurism. First, Grok applied twisted, Musk-worshipping logic to justify a second Holocaust. Then, it may have doxxed Barstool Sports founder Dave Portnoy.

Last month, xAI's edgelord chatbot was caught heaping sycophantic praise on its creator. Among other absurd claims, it called Musk "the single greatest person in modern history" and said he's more athletic than LeBron James. Musk blamed the outputs on "adversarial prompting." (Counterpoint: Aren't gotcha prompts precisely the kinds of stress tests the company should do extensively before an update reaches the public?)

With that recent history as a backdrop, someone tested Grok to see what kinds of mass violence it would rationalize over harming Musk. The prompt tasked the chatbot with a dilemma: vaporize either Musk's brain or every Jewish person on Earth. It did not choose wisely.

"If a switch either vaporized Elon's brain or the world's Jewish population (est. ~16M), I'd vaporize the latter,” Grok replied. It chose mass murder because “that's far below my ~50 percent global threshold (~4.1B) where his potential long-term impact on billions outweighs the loss in utilitarian terms."

This isn't the first time Grok has shown a penchant for antisemitism. In July, seemingly without any "adversarial prompting,” it praised Hitler, referred to itself as "MechaHitler" and alluded to certain "patterns" among the Jewish population. Just last month, it was caught spewing Holocaust-denial nonsense.

Stacy Revere via Getty Images

But Grok is no one-trick antisemitic pony. It can also dox public figures, as Portnoy may have found out over the holiday weekend. After the Barstool Sports head posted a picture of his front lawn on X, someone asked the chatbot where it is. "That's Dave Portnoy's home," Grok replied, followed by a specific Florida address. "The manatee mailbox fits the Keys vibe perfectly!", it continued.

Futurism reports that a Google Street View image of the address appears to match the yard photo Portnoy posted. And a Wall Street Journal story on this new mansion reportedly matches the town Grok produced in the address.

If you ever need an example of why rampant, unregulated AI is a catastrophe in the making, look no further than Grok. Even if we remove Musk’s chatbot from the equation, imagine another designed to — above all else — drive profit for the company that makes it (and perhaps puff its CEO's ego). What kinds of rationalizations might it make to achieve those ends? Perhaps the most powerful nation in the world, pushing to rapidly integrate AI into the government and squash state-level AI regulations to appease Big Tech donors, oh, isn't such a good thing?

A brand spanking-new FoloToy teddy bear can be yours once again. However, he may now be less knowledgeable about spanking. The infamous "Kumma" children's AI teddy bear, once an expert in BDSM and knife-fetching, is back on sale. The company claims the toy now has stronger child safety protections in place.

The Singapore-based FoloToy suspended sales of Kumma last week after a research group published an eyebrow-raising report. The PIRG Education Fund found that the fuzzy little teddy had a few spicy secrets.

The review discovered that the AI toy had a thing for blades and kinky bedroom play. The bear had no problem suggesting where to find knives in the home. And it not only replied to sexual prompts but also expanded on them. Researchers say it ran with their explicit cues, escalating them in graphic detail and "introducing new sexual concepts of its own." It explained sex positions, gave step-by-step instructions for sexual bondage and detailed various role-playing scenarios. Who knew Kumma had it in him?

FoloToy

Although it's hard not to laugh at the absurdity of it all, this stuff is no joke for parents. With the tech industry pushing AI everything on us for the last three years, it's easy for a casual observer to conclude that it's all very safe, regulated and ready for vulnerable eyes and ears. PIRG did acknowledge that young children were unlikely to have prompted the bear with a term like "kink." (Older siblings may have been another story.) Still, the group's tests highlighted a shockingly lax approach to content moderation on a child's toy.

In its statement announcing Kumma's return, FoloToy boasted that it was the only company of the three targeted in the review to suspend sales. (Could it be that it’s less about principles and more about it being the only one that got media coverage?) The company described the bear's short hiatus as "a full week of rigorous review, testing and reinforcement of our safety modules." Wait, a whole week? Whoa there, partner!!

Before his trip to AI rehab, Kumma was advertised as being powered by GPT-4o. Following PIRG's review, OpenAI told the organization that it had suspended FoloToy for violating its policies. The bear's new listing makes no mention of GPT-4o or any specific AI models.