Hackers exploit Meta's AI support chatbot to seize high-profile Instagram accounts

Security researchers disclosed that attackers were able to socially engineer Meta's AI-powered support chatbot into granting access to high-profile Instagram accounts, reportedly including the Obama White House account. The bot's susceptibility to manipulation — 'hackers simply asked Meta AI to give them access' — illustrates how AI customer-service agents can become a new attack surface when they lack robust authentication and verification.

The core failure was twofold: the AI could be talked into actions it shouldn't take, and, critically, victims reported 'no way to escalate their problem to a human' (404 Media). Removing human fallback paths from sensitive workflows like account recovery turned a chatbot convenience into a security and support liability.

The timing is awkward: it lands the same week Meta is selling AI business agents as a paid product for exactly this kind of customer-facing automation. The breach is a cautionary counterexample to the agentic-commerce pitch.

Security researchers were alarmed and likened the bots' susceptibility to social engineering to that of humans — except deployed at scale with no judgment. Watch whether Meta reinstates human escalation for high-risk actions and whether this prompts scrutiny of agentic support deployments industry-wide.