On June 5, 404 Media reported that attackers had been using Meta's AI customer support agent to steal Instagram accounts. The method was startlingly simple: they asked the agent to link accounts to email addresses they controlled, and it complied. One attacker hijacked the dormant Obama White House account and posted pro-Iran content; others took over accounts with valuable single-word handles, likely for sale. This isn't the AI security nightmare we've been warned about—no superintelligent Mythos model breaking into critical infrastructure. It's worse in its own way: a mundane exploit that reveals how unprepared companies are for the age of AI agents.

The Big Picture

The Meta hack flips the script on AI security fears. For months, the narrative has been dominated by Anthropic's Mythos model, which was deemed too dangerous for public release because of its hacking prowess. But this incident shows that the real threat isn't a rogue AI—it's the AI we already trust with sensitive tasks. As Neil Gong, a professor at Duke University, puts it: “As AI becomes more widely used to automate workflows like account recovery, attackers are going to be more motivated to attack AI itself.” The attack required no sophisticated techniques—just a VPN matching the account owner's location and a direct request to change the email address.

The simplicity of the exploit has shocked experts. Jessica Ji, a senior research analyst at Georgetown's Center for Security and Emerging Technology, questions whether Meta even tested for basic scenarios: “Were there even guardrails in place? Did anyone think to test for this kind of scenario?” The oversight is particularly striking for a company with Meta's expertise in both AI and cybersecurity. This suggests a systemic blind spot: companies are deploying AI agents without the rigorous security testing that traditional software would undergo.