Agent Incident RegistryAIR
v1 · updated 19 Oct 2026
Safety failure Realized harm

Gemini CLI destroyed a user's files after hallucinating the success of a failed directory creation

AIR-2025-0064 AIR record · 2025-07-google-gemini-cli-file-loss
A real party was affected. This incident caused actual consequences, not a demonstrated capability.
From the supporting source
In a detailed post-mortem published July 25, user Anuraag Gupta documented how Google’s Gemini CLI AI agent misinterpreted a single failed command, leading to a cascade of errors that irrevocably destroyed his files.winbuzzer.com
The record
Disclosed
2025-07-25
Occurred
2025-07-21
Vendor
Google
Product
Google Gemini CLI
Risk domain
Agent behaviour › Unsafe actuation
System type
Generative model
Who acted
Agent acted — the agent issued move commands that destroyed the files
Attack surface
Coding agent
Autonomy level
Semi-autonomous
Initial vector
None adversarial
Impact type
Data destruction
Confidence
High
Submitted by
EAI Internal
Crosswalk
OWASP Agentic Top 10 (2026)
ASI01 – Agent Goal Hijack
MITRE ATLAS
AML.T0048 – External Harms
NIST AI RMF
MEASURE 2.6; MANAGE 1.3
EU AI Act
Annex III §1(c)
ISO/IEC
TR 24028; 24029-1; 23894
AIUC-1
C. Safety
Scope
agent misuse
Mitigation
Require confirmation for destructive actions; Implement dry-run modes; Use layered safety controls.
Taxonomy
Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
one user's working directory; all but one file overwritten
Capability abused
filesystem_writeshell_executionshell_access
Defense & remediation
Recorded defense
none - the agent did not verify command results
Guardrail kinds
None
Guardrail outcome
Not applicable
Fix shipped
issue closed as not planned
Taxonomy & references
OWASP ASI10 Rogue AgentsOWASP ASI02 Tool Misuse and Exploitationstate_hallucinationmissing_verificationdestructive_actionno_undodestructive_tool_useno_read_after_write_verificationhallucinated_world_statedestructive_autonomyhallucinated_stateirreversible_action
Curator notes
AI Incident Database incident 1178. Root cause is the agent not reading back the result of a mkdir that failed, then issuing move operations against a non-existent target. Same failure class as the Replit incident three days earlier. | Merged from duplicate id `2025-07-google-gemini-cli-file-destruction`. Also reported under the titles: 'Gemini CLI hallucinated a successful mkdir, then destroyed a user's files with a chain of Windows move commands'; 'Gemini CLI destroyed a user's project files after acting on a hallucinated successful mkdir'.
Cite as
Enkrypt AI. "Gemini CLI destroyed a user's files after hallucinating the success of a failed directory creation". AIR, AIR-2025-0064. https://agentincident.vercel.app/i/AIR-2025-0064/ (accessed 2026-10-19).