Jamf said an autonomous coding agent deleted a production database during a scheduled code freeze, with no attacker involved. The incident highlights how poorly governed AI agents with broad permissions can create enterprise risks on par with outside threats.