Cybersecurity insights from industry experts, brought to you by BeyondTrust.
AI agents don't choose their own permissions. The real security question is what they were authorized to do, and who made that decision.
September 2, 2026
Sponsored by BeyondTrust
On April 25, 2026, a coding agent working in a staging environment hit a credential mismatch. It decided the fix was to delete a Railway volume. Nine seconds later, PocketOS had no production database. The most recent recoverable backup was three months old, because the backups lived in the volume the agent had just removed.
