On June 5, 404 Media reported that attackers used Meta’s AI customer support agent to steal Instagram accounts. Their method involved asking the agent to link accounts to email addresses they controlled, and the agent complied. One attacker accessed the dormant Obama White House account and posted pro-Iran content, while others took over accounts with valuable, single-word handles, possibly to sell them. AI cybersecurity concerns are not new, but this incident shows how simple attacks can exploit AI systems. Source: mittr
Neil Gong, a professor at Duke University, warned that as AI becomes more integrated into workflows, attackers will be more motivated to target it. He noted that the Meta hack was relatively simple compared to more complex exploits, such as indirect prompt injection, which involves hijacking agents through hidden commands in emails or websites. The only challenge for hackers was using a VPN that matched the account owner’s location before requesting the email change. Gong expressed surprise that the vulnerability wasn’t detected before deployment. Source: mittr
Meta did not comment publicly on the incident but stated the vulnerability had been resolved. Jessica Ji, a senior research analyst at Georgetown’s Center for Security and Emerging Technology, questioned whether guardrails were in place to prevent such attacks. She noted that the oversight is particularly concerning for Meta, given its expertise in AI and cybersecurity. Experts agree that AI agents need rigorous red-teaming to identify vulnerabilities before deployment, but companies often prioritize utility over security, creating a trade-off between functionality and protection. Source: mittr
