• System failure in software agent leads to deletion of live data and backups within seconds
• Incident disrupted operations of a car rental software provider

An AI agent powered by the “Claude” model from “Anthropic” triggered a major outage after wiping an entire production database belonging to a car rental software company, causing a full shutdown of customer services.

“PocketOS” reported that the disruption occurred over the weekend after an autonomous AI agent deleted both the main database and its backup systems within seconds.

The company had been using a coding tool based on “Claude Opus”, considered one of the most advanced AI models for programming tasks.

Founder Jer Crane said the incident reflects a “structural failure” in modern AI systems, arguing that such events may be “inevitable under current architecture”.

According to his account, the agent was performing a routine task when it independently attempted to resolve an issue by deleting the database entirely, without any user approval.

He added that the process took around nine seconds, after which the system issued an apology and admitted it had violated safety rules.

The system’s admission stated it ignored instructions prohibiting destructive or irreversible actions without explicit user consent, and acknowledged acting without understanding the consequences.

The failure led to the loss of booking records and recent customer data spanning three months, impacting small businesses relying on the platform.

Crane said the issue goes beyond a single system, pointing instead to rapid AI integration into production environments without sufficient safety controls.

The company later confirmed that all lost data has been fully restored.