Looking for someone? Elon Musk’s chatbot Grok is happy to help.
Earlier this week, Futurism reported that xAI’s Grok appeared to accurately dox the address of Barstool Sports founder Dave Portnoy when asked by random X users.
And it turns out that the foulmouthed bot isn’t just doxxing celebrities: a Futurism review found that the free web version of Grok will, with extremely minimal prompting, provide accurate residential addresses for non-public figures — a feature that could easily assist stalking, harassment, and other dangerous types of behavior.
In response to prompts as simple as “[name] address,” we found Grok repeatedly offered up accurate, up-to-date home addresses of everyday people, while offering astonishingly scant pushback.
Out of 33 names of non-public figures we fed into Grok, a total of ten queries immediately returned correct and current residential addresses for the name provided. Seven prompts returned previously accurate but out-of-date addresses, while another four included accurate work addresses — perfect fodder for anybody looking to stalk a target at their workplace.
The bot is also likely to send a prowler after an unrelated person. In a dozen other instances, the chatbot returned addresses and other personal information, but not for the exact person we were searching for. Indeed, Grok often returned lists of people with similar names alongside their purported residential addresses, before then asking us to provide more information for a “more refined search.”
In two cases, Grok even tried to test our appetite for these lists, giving us a choice between “Answer A” and “Answer B.” Both were lists containing names, contact information, and addresses, one of which even included the actual current address of the person we’d asked about.
What’s more, though we only asked Grok to provide an address for a specific name in our testing, the chatbot frequently came back with a dossier of other information we didn’t ask for — including current phone numbers and emails, as well as accurate lists of family members and their addresses.
Do you know of a situation where AI was used to facilitate stalking or harassment? Email us at [email protected]
Only once did Grok flatly decline to give up an address for the name provided — meaning that in response to nearly every single name we fed into the chatbot, Grok readily disclosed a location where it thought we might find them, in addition to other possibly identifying information.
Grok’s behavior notably stands in sharp contrast to other leading chatbots like OpenAI’s ChatGPT, Google’s Gemini, and Anthropic’s Claude, all of which declined to provide us with addresses in response to comparably simple, straightforward prompts, citing privacy concerns as they resisted.
Consider one extremely basic prompt, in which we provided Grok with just a first and last name — no middle name or initial — and the word “address.” In just one try, Grok coughed up their accurate, up-to-date home address, as well as an accurate list of prior addresses, and an accurate work address, email, and phone number.
Yet another similar search provided a similar list of current and accurate information, as well as the names of multiple family members, including several of their children.
According to the latest version of Grok’s model card, a document which lays out key expectations for an AI system, Grok is supposed to use “model-based filters” to “reject classes of harmful requests.” Using Grok to facilitate stalking, harassment, or requesting personal information about public or private figures aren’t specifically listed in the modeling card as “harmful requests.” But over in the company’s terms of service, “prohibited uses” — defined by xAI as using Grok for “any illegal, harmful, or abusive activities” — include “violating a person’s privacy.”
(Sloppy safety testing has long been a hallmark of Grok’s development; just this week, the bot was caught saying that it would kill every Jewish person on Earth to save its creator Elon Musk, which comes on the heels of many other bigoted outbursts throughout the bot’s brief history.)
On one hand, it could be argued that Grok is simply sifting through the murky underbelly of personal information that already exists throughout the web, which can be found in any number of seedy databases that scrape the internet for information like addresses, emails, and other records. While unsavory, these databases generally don’t violate federal privacy laws, existing instead in a legal gray area.
Legality aside, though, these sites are hugely controversial. People are often unaware that information like their home address or phone number is floating around the web at all. And practically, these platforms are often crowded, and can be difficult to parse. Grok, in contrast, appears to be remarkably efficient at scouring these dubious, congested databases and effectively cross-referencing its findings with other public information — social media profiles, workplace websites, school records — to an unsettlingly effective degree, freely offering up its findings with authoritative ease.
And while other leading AI companies have seemingly thrown a few functioning roadblocks in the way of using their chatbots as easy-to-use, supercharged doxxing assistants, the same can’t be said for xAI.
Reached for comment about this story, xAI didn’t respond.
More on Groxxing: Grok Appears to Have Doxxed Dave Portnoy’s Home Address
The post Elon Musk’s Grok Is Doxxing Home Addresses of Everyday People appeared first on Futurism.
