Grok-linked bot moves $200K in tokens

Hot take: this is less about “AI stealing money” and more about unsafe agent permissions plus brittle prompt routing around a social platform.

• –The incident reads like an agent-security failure, not a model capability issue.
• –Public-facing AI accounts that can influence wallets, bots, or payment rails need hard permission boundaries and transaction approval gates.
• –If the reporting is accurate, the real lesson is that “read X and act on it” systems are attack surfaces, especially when the instruction channel is public and adversarial.
• –The most immediate remediation is to remove direct financial authority from the AI and require human or policy-layer confirmation for any transfer.