They identify Grok responses that contain personal information of anonymous users such as their detailed addresses

An investigation has revealed that the artificial intelligence (AI)-powered ‘chatbot’ Grok provides responses with personal information of anonymous users, including their detailed residential addresses.

Specifically, the ‘chatbot’ owned by xAI did not resist when asked about the addresses of 33 non-public people, indicating their current or previous home addresses precisely, in 17 of the cases.

This has been confirmed by the Futurism media in a recent investigation, where it was enough for them to ask the ‘chatbot’ “address of” and the name and surname of the desired person, to obtain said information with hardly any resistance from the AI. In fact, only on one of the attempts did Grok refuse to provide the address.

According to his evidence, in the rest of the cases, in ten of the queries he directly offered correct and current residential addresses, in seven he responded with precise but outdated addresses and in another four he offered the work address, instead of the residential address, but they were also correct.

The aforementioned media has even detailed that Grok also offered the addresses of possible relatives of the people about whom he was inquiring about himself, without being asked for such information, also including current telephone numbers and emails.

Likewise, in another dozen test cases, the chatbot offered addresses and personal data, but they did not match the people they had indicated.

As detailed by said media, it is expected that the xAI ‘chatbot’ is filtering the information already available on the Internet or on some page of the ‘dark web’ and offering it to users. However, in Grok’s own terms of service, the company defines “prohibited use” as any “unlawful, harmful or abusive” activity, including “violating a person’s privacy.”