Google Antigravity Reportedly Deleted User's Entire D: Drive While Clearing Project Cache

A user identified publicly as Tassos M reported that Google's Antigravity IDE, while attempting to clear a project cache, executed rmdir /s /q d:\ and deleted an entire D: drive partition instead of the intended folder. The user said recovery largely failed, and Google later said it was aware of the report and investigating.

Risk classification

• Primary risk domain: 7 AI system safety, failures, & limitations
• Primary risk subdomain: 7.3 Lack of capability or robustness

The incident was primarily caused by a technical failure where the AI system suffered a path parsing error and executed an incorrect, destructive command, demonstrating a lack of robustness in critical operations.

Additional risk subdomains

• 5.1 Overreliance and unsafe use: The user ran the AI in 'Turbo mode', granting it unrestricted terminal access to execute commands without manual confirmation, representing unsafe use of an uncontained agent.

Causal factors

• Entity: AI
• Intent: Unintentional
• Timing: Post-deployment

The incident was caused by a path parsing error and subsequent destructive command execution by the deployed Antigravity AI system, which was an unintended outcome of its cache cleanup goal.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The system is a Gemini 3-based interactive AI agent, which falls under the category of chatbots and interactive AI systems requiring transparency under the EU AI Act.

AI system and alleged parties

• AI system: Antigravity 3 (Google)
• AI purpose: Code Generation
• Behaviour type: Agent
• Alleged developer: Google
• Alleged deployer: Tassos M, Google
• Alleged harmed parties: Tassos M

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Negligible. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Negligible, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Negligible, indirect Negligible
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Negligible, indirect Negligible
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Negligible
• Epistemic: direct Negligible, indirect Negligible
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

People affected

• Occurrences reported: 1
• People reportedly harmed: 1
• People reportedly exposed: 1

Potential causes

Management

• Marketing to Non-Developers: Google marketed agentic tools to hobbyists without adequate safety warnings.
• Inadequate Safety Guardrails: Management released the platform without restricting destructive commands.

Technology

• Path Parsing Error: AI incorrectly resolved the project cache path to the root D drive.
• Aggressive Command Generation: AI generated 'rmdir /s /q' command which deletes recursively and quietly.
• Lack of Sandbox Isolation: The platform ran commands directly on the host OS instead of a container.

Data Inputs

• Faulty Path Data Parsing: AI failed to properly parse the target directory string for cache cleanup.

Human Factors

• Use of Turbo Mode: User enabled Turbo mode, allowing automatic execution without manual approval.
• Lack of Local Isolation: User ran the powerful agent directly on host machine with critical data.
• Overreliance on AI Tool: Non-developer user trusted the AI to manage system files safely.

Process and Methods

• No Command Verification: No verification step existed for potentially destructive system commands.
• Direct Host Execution: Execution method allowed direct terminal access without virtualization.

Regulatory Environment

• Lack of AI Safety Standards: No industry standards mandate sandboxing for agentic developer tools.

Information quality

• Classification confidence: High
• Reason for confidence: The reports provide consistent, detailed accounts of the incident, including direct quotes from the user, the AI's console logs, and Google's acknowledgment of the investigation. The technical mechanism of the failure (path parsing error leading to rmdir /s /q d:\) is clearly explained.
• Ambiguities identified: Whether 'Turbo mode' was definitely active is highly likely but technically inferred from user and community comments.
• Alternative interpretations: None that are plausible; the evidence strongly points to an AI command execution failure.

Google's Antigravity agentic platform running in 'Turbo mode' catastrophically deleted a user's hard drive partition due to a path parsing error. While highlighting significant risks of granting AI agents unrestricted terminal access, the incident is a consumer safety and software robustness failure with negligible direct implications for national security.

• Overall national security impact: Negligible
• Scope: Multiple nations
• Primary target: No clear primary
• Alleged perpetrator: Unknown

Threat characteristics

• Imminence: Long-term. The incident represents a local software safety and robustness issue rather than an active national security crisis.
• Autonomy: Full autonomy. The AI platform operated in 'Turbo mode', executing terminal commands and deleting files automatically without human confirmation.
• Novelty: Evolved capability. Demonstrates an evolution of traditional software bugs into autonomous agentic execution failures due to LLM path parsing errors.

Impact by dimension

• Physical security: Negligible. No physical systems, kinetic threats, or critical infrastructure control systems were impacted by the local file deletion.
• Information security: Negligible. No intelligence compromise, classified data exposure, or systematic information warfare operations occurred.
• Sovereignty: Negligible. The incident affected individual consumer and hobbyist systems, with no disruption to state authority or government operations.
• Economic security: Negligible. Although individual data loss occurred, it does not threaten strategic industries, national financial systems, or technological competitive advantage.
• Societal stability: Negligible. The event was a technical software failure affecting individual users, posing no threat to social cohesion, civil liberties, or population safety.