This would irritate us, too.
Estoy Aqui
Meta’s AI chatbot kept giving strangers a man’s phone number, and it remains unclear how or why it happened.
As Business Insider‘s Rob Price reports, the debacle began when he was added to a random WhatsApp group chat filled with people from multiple countries who began asking him random questions in Spanish.
The people in the chat seemed bemused when the senior Silicon Valley correspondent asked them why he’d been added to the group chat and why they were asking him questions. Eventually, someone posted a screenshot and Price noticed that his number was saved under the name “Meta AI.”
After more back-and-forth, someone eventually shared another screenshot that included a one-on-one exchange with Meta AI, the company’s cross-platform chatbot, explaining that Price’s phone number could be used to add it to group chats.
“You can add me to a WhatsApp group as if I were just another contact,” the chatbot told the user in a translation of the original Spanish exchange. “You only need to save my phone number.”
After that first bizarre instance, people from various countries in South America kept contacting the reporter thinking he was Meta AI, spitting out Price’s work number each time when users asked the chatbot for its phone number.
Working Theory
As a journalist, Price was accustomed to being contacted by random people, but this was substantially different. Generally speaking, being accused of being a bot only happens during particularly intense political debates between humans on social media — and this was nothing of the sort.
Ultimately, the reporter concluded that when training the large language model (LLM) undergirding the Meta AI chatbot, the company had likely scraped his publicly-available phone number, too.
Having written roughly 300 stories mentioning Facebook that included his phone number — which was shared, as is common practice, to solicit tips — Price reckons that the number may have been “scooped up” into Meta’s training data. During the mysterious “black box” sausage-making process that results in AI, the LLM may then have made some “misguided causal connection” that resulted in his number being spat out when users asked Meta for its phone number.
In a statement to BI, a company spokesperson seemed to co-sign this theory, telling Price that because Meta AI “was trained on publicly available information online,” his number may have ended up in the secret sauce.
The only problem? Axel Springer, Business Insider‘s parent company, doesn’t have any deal in place to allow Meta to train its AI on its content (though it does have one with OpenAI).
Price noted that after he’d contacted Meta about the bizarre occurrences, people stopped randomly messaging him thinking he was a chatbot — and that he’d never been able to replicate the fluke for himself, either.
More on Meta AI: Facebook Is Euthanizing All Its Celebrity AI Clones
