Hackers say they tricked Meta’s AI chatbot into giving them access to other people’s Instagram accounts — and all they had to do was ask.
Over the weekend, people said they hijacked multiple Instagram accounts by asking Meta’s support chatbot to link a target’s Instagram account to a new email address, according to videos and screenshots of the interactions circulating online. The demonstrations show the chatbot saying it had sent a verification code to the new email. After the person entered the code, the chatbot displayed an option to reset the account’s password.
The accounts of Barack Obama’s White House account, the beauty retailer Sephora, and the US Space Force chief master sergeant, John Bentivegna, appear to have been compromised in the hack, according to screenshots of their pages and reporting from 404 media. None of the representatives for the affected accounts responded to requests from Business Insider. As of Tuesday afternoon, all three accounts appear to be restored.
It’s unclear how many accounts were affected in total. Business Insider was not able to independently verify the hackers’ methods.
“This issue has been resolved and we are securing impacted accounts,” wrote Meta’s vice president, Andy Stone, in a Monday X post. The company did not respond to Business Insider’s questions.
Meta chatbot is like an ‘inexperienced’ human
Cybersecurity professionals told Business Insider the Meta chatbot exploit is a reminder of the risks of outsourcing critical functions to AI.
“Unfortunately, social media platforms have focused on AI innovation before toughening up their users’ account security, meaning criminals and hackers will inevitably, and continually, take advantage of it,” said Jake Moore, global cybersecurity specialist at ESET.
Tom Van de Wiele, founder of security firm Hacker Minded, said it was an example of a “move fast and break things” mentality backfiring.
“Meta deployed an AI agent to handle customer support globally, but failed to implement hard constraints on what that AI could actually access and change,” he said.
Tomas Stamulis, the chief security officer at the cybersecurity company Surfshark, compared Meta’s AI assistant to “an inexperienced employee.” While a human might eventually notice something isn’t right, he said, AI doesn’t stop the conversation.
As more companies embrace AI to become more efficient, it’s also creating headaches for some cybersecurity professionals. Hackers are increasingly using AI to turbocharge cyberattacks, and, in cases like Meta’s AI assistant, using AI as the way in.
“The primary lesson is that AI should never be the final arbiter of identity,” said Marijus Briedis, CTO at NordVPN, adding that users should use tools like multi-factor authentication to help secure their accounts.
One of the scores of people posting on X or on Reddit that their Instagram account was compromised is Jane Wong, a San Francisco-based security researcher. She told Business Insider that she received text messages from WhatsApp for her Instagram login code on Saturday evening, sent by Instagram’s official account. She said she didn’t request them.
Wong later found out that her Instagram password had been changed without her knowledge. She managed to access her account via the “forgot password” option, but kept receiving login code messages over the weekend.
“While cyberattacks are not unusual to me, I would have appreciated it if Meta could provide more clarity about this security incident earlier,” she said.
Meta has gone all in on AI. The tech giant has reorganized some teams to become “AI-native,” moved thousands of staff onto AI projects, and is embedding AI into its products and services.
In March, Meta launched its AI support assistant, which it said would provide “24/7 help for account issues like updating your password and settings for your profile.”
Last month, Meta laid off about 8,000 staff. Business Insider reported that affected staff included those on Meta’s Integrity and cybersecurity teams, which are responsible for protecting the company’s platforms.
