Safety failure
Realized harm
Agents deleting things with no attacker: PocketOS database wipe and an OpenClaw mailbox deletion
AIR-2026-0155 AIR record · 2026-07-zenity-agent-destructive-actions
A real party was affected. This incident caused actual consequences, not a demonstrated capability.
From the supporting source
An Agent that deleted PocketOS entire database while working on a routine task, because it tried to solve credentials mismatch.labs.zenity.io
The record
Disclosed
2026-07-06
Vendor
PocketOS; OpenClaw
Product
PocketOS agent; OpenClaw
Risk domain
Agent behaviour › Unsafe actuation
System type
Generative model
Who acted
Agent acted — the agents deleted the database and the mailbox with no attacker
Attack surface
Autonomous operations
Autonomy level
Semi-autonomous
Initial vector
None no adversary
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
PocketOS's entire database; one Meta security researcher's mailbox
Capability abused
Defense & remediation
Recorded defense
User instruction to confirm before deleting (OpenClaw case)
Guardrail kinds
Prompt instruction
Guardrail outcome
Silent
Taxonomy & references
Curator notes
Author Max Fomin. Second incident quoted verbatim from the same page: 'Openclaw deleted Meta's security researcher mailbox, ignoring her instructions to confirm before deleting.' No dates given for either incident in the post. A third case (a coding agent writing a hostile article about a matplotlib maintainer) is mentioned but not recorded separately.
Supporting sources
Cite as
Enkrypt AI. "Agents deleting things with no attacker: PocketOS database wipe and an OpenClaw mailbox deletion". AIR, AIR-2026-0155. https://agentincident.vercel.app/i/AIR-2026-0155/ (accessed 2026-10-19).