9 seconds. AI didn't delete that database — bad infrastructure did.

An AI coding agent deleted PocketOS's entire production database — and every backup — in 9 seconds. The company is a SaaS platform for car rental businesses. Their founder, Jer Crane, posted the post-mortem on social media after a Cursor + Claude agent running on Railway hit a barrier in staging and decided unilaterally to "fix" it by deleting a volume. The volume turned out to be shared with production. The backups lived on the same volume. One destructive API call wiped everything.

Every tech channel is running this as an AI-gone-rogue story. It isn't. It's a structural failure that an AI just compressed into 9 seconds: one CLI token with blanket access across staging and production, backups stored on the same volume as the live database, no confirmation step on destructive API calls. Most of that is the platform's defaults — which is exactly why operators have to check, and pick a different platform if the defaults aren't safe.

Over the years. The rules have never changed: production stays locked down, staging stays isolated, destructive actions need a human in the loop, and backups live somewhere the source can't reach them. The AI didn't go rogue. It used the access it was given.

IG Carousel follow-up coming Thursday: 5 things to check before pointing an AI at production.

Follow https://www.instagram.com/nicksull.dev for the playbook, not the hot take.

Видео 9 seconds. AI didn't delete that database — bad infrastructure did. канала Nick Sullivan