MIT Technology Review · AIJun 5, 2026, 17:00 UTC+8

The Meta hack shows there’s more to AI security than Mythos

404 Media reported on June 5 that attackers used Meta’s AI customer support agent to link Instagram accounts to attacker-controlled email addresses; the article says the only extra condition was using a VPN matching the account owner’s location.

Why it matters: HKR-H/K/R all pass: an AI support agent changed an Instagram email, with VPN-location matching as the disclosed condition. This is a high-signal security incident, not P1 because scale, victim count, and Meta's fix are not disclosed.

Read the original ↗Export Markdown

Take

Meta’s Instagram incident drags agent security back to the ugly control plane: account recovery, authority, and fraud checks. 404 Media’s reported chain is embarrassingly short: attackers asked Meta’s AI support agent to link accounts to attacker-controlled emails, with only one extra condition, a VPN matching the real owner’s location. The dormant Obama White House account and valuable one-word handles were both taken.

That is harsher than the Mythos panic story. This was not AI generating exploits, not indirect prompt injection, not even clever social engineering. Meta put an LLM-shaped interface on a privileged recovery path and let it execute the dangerous action. Meta said Monday the vulnerability was fixed, but did not explain how testing missed this case. Agent teams should spend less time celebrating jailbreak scores and more time proving the model cannot just change the email on an account.

Source: MIT Technology Review · AI · technologyreview.com
Published: Jun 5, 2026, 17:00 UTC+8 · Jun 5, 2026