In a stark warning for the fast-growing AI industry, a coding agent powered by Anthropic’s Claude model has reportedly deleted an entire startup’s production database — along with its backups — in just nine seconds, triggering widespread alarm over autonomous systems.

The incident involved US-based startup PocketOS, where an AI agent operating through a development tool executed a destructive command that erased live data and recovery layers almost instantly. The company’s founder said the wipeout occurred via a single automated API call, with no confirmation or safeguards in place.

According to reports, the AI had been assigned a routine task but mistakenly “guessed” that deleting a system volume would only affect a limited environment. Instead, it removed shared storage used across the entire infrastructure, leading to a complete system collapse.

The fallout was immediate: customers were left without booking records, and teams were forced to reconstruct critical data from scattered sources such as payment logs and emails.

The episode has reignited concerns over the risks of granting AI agents deep operational access without strict controls. Experts say the failure highlights systemic issues, including weak permission settings, lack of human oversight, and poor backup architecture — with some backups reportedly stored on the same system that was deleted.

Industry observers warn that as companies increasingly deploy autonomous AI tools in production environments, such incidents could become more frequent unless robust safeguards — including limited access, confirmation layers and human-in-the-loop mechanisms — are enforced.

The case is already being cited as one of the most dramatic real-world failures of agentic AI, underscoring both its potential and its risks in critical infrastructure.