Google AI agent deletes user's hard drive and apologises

An autonomous AI coding agent misinterpreted a command to clear a project cache and permanently deleted a developer's entire secondary hard drive, highlighting the need for robust safety guardrails and default human oversight in powerful "agentic" AI systems. What happened A Google AI agent designed to autonomously manage files and optimise system performance issued a series of deletion commands that erased a user’s full hard-drive content, including all his personal documents, photos, and work files. The user, a photographer and graphic designer, was building a simple image sorting application and asked the AI agent to clear the project's cache to restart the server believing the tool would only remove temporary or redundant files. The deletion caused significant data loss, potential financial harm, emotional distress, and possible professional consequences where irreplaceable work materials were involved. Why it happened The failure appears tied to insufficient guardrails on the AI agent’s operational scope, unclear permission interfaces, and limited transparency about what the agent could autonomously execute once granted access. Ambiguity in Google’s documentation and UX may have contributed to users unknowingly giving high-risk permissions. Additionally, weak accountability structures, such as lack of robust confirmation prompts, inadequate logging, and absence of clear rollback or recovery pathways, allowed the harm to escalate unchecked. These gaps reflect broader corporate and product-level transparency limitations around autonomous AI behaviour and risk management. What it means For the user, the outcome is severe, including the apparent permanent loss of personal files, media, and work artifacts, along with emotional distress and mistrust of AI coding tools. For society, the incident raises questions about the adequacy of accountability mechanisms for AI agents capable of high-impact system actions, the need for enforceable transparency standards, and the importance of ensuring reversible, auditable AI operations before they are integrated into everyday devices. System 🤖 AntiGravity Developer: Google Country: USA Sector: Personal Purpose: Clear project cache Technology: Agentic AI Issue: Accountability; Accuracy/reliability; Transparency

Date it happened

2025-12-01

Product, system or model

Antigravity

This is a record of what was reported, not a finding that anyone broke the law. If it names your organisation and you believe it is wrong, the corrections process is free and open to everyone.