Meta's AI chatbot helped attackers take over Instagram accounts

An AI chatbot from Meta could be used by third parties with simple prompts to take control of any account on Instagram. Details have now become known after, among other things, an old account of the White House under Barack Obama shared Iranian propaganda. Whether there is a connection is unknown. The attack vector was made public by US tech magazine 404 Media, which refers to Telegram groups of security researchers and hacking groups. The Instagram company Meta has already reacted; the company spokesperson assured on X: "This issue has been resolved and we are securing impacted accounts."

Shockingly Simple Method

According to the report, to take over an account, one only had to pretend via VPN that they were in the approximate home region of the person whose Instagram account was to be hijacked. Next, an AI chatbot could be prompted for support requests to connect the desired account with a new email and then send a password reset code there. The chatbot complied, 404 Media reports. How often such attacks were actually successfully carried out is unclear. However, on the microblogging service X, the well-known app researcher Jane Manchun Wong explained that her account password was changed using this method.

In March, Meta launched an AI assistant for support requests, which was supposed to help with problems on Facebook and Instagram around the clock. The company stated at the time that the AI should help to "update your password and change settings in your profile." The AI assistant was designed to solve problems with an account and not just make suggestions, it was said at the time. 404 now writes that long lists of account names, allegedly taken over by exploiting this feature, have been circulating on Telegram. However, this cannot be verified. The attack method shows the enormous risk associated with outsourcing technical support to AI technology, the portal further writes.

(mho)