Skip to content
Submit incident
AII-1000-59097
ConfirmedA court, a regulator or the organization itself has stated the core facts, or the record cites a court or regulator source.Supporting sources are pending verification.

Hong Kong Privacy Commissioner warns of agentic AI risks

Hong Kong's Privacy Commissioner issued a warning about security and privacy risks from agentic AI systems, particularly OpenClaw, before deployment. The office recommended limiting system permissions granted to such systems and avoiding sharing sensitive personal data with them.

16 March 2026 Β· Office of the Privacy Commissioner for Personal Data, Hong Kong Β· Safety and wellbeing
16 March 2026
AII-1000-59097
Quick facts
Category
Safety and wellbeing
Country
Hong Kong
Date
16 March 2026
Drafted by
Raidu Incident Detection

What happened

The Hong Kong Office of the Privacy Commissioner for Personal Data issued a warning about security and privacy risks from agentic AI systems, particularly OpenClaw, advising organizations and members of the public to be cautious before deployment and use.

The office identified risks including potential data leakage, malicious system hijacking, and the possibility of misinterpreting user instructions leading to unintended deletion of important data.

Reported impact

The Privacy Commissioner noted that agentic AI poses higher privacy risks compared to standard AI chatbots due to the extended capabilities and permissions these systems require to operate.

Index analysis

The office recommended that agentic AI systems should be granted only the minimum permissions necessary to complete assigned tasks, and that users should avoid providing sensitive personal data such as identity documents, bank account numbers and passwords.

The Privacy Commissioner advised users to retain final control over data transmission and system configuration changes, and to verify that plugins use official latest versions free from malicious code.

Governance controls

The Privacy Commissioner recommended that organizations should only grant agentic AI systems the minimum permissions required to complete tasks and should avoid granting administrator account permissions.

The office advised users not to readily provide personal data to agentic AI systems and to maintain ultimate control over data transmission and system configuration modifications.

Classification

Category
Safety and wellbeing
Industry
Digital security
Country
Hong Kong
Harm status
Potential harm
Intent
Unintentional
Timing
Pre-deployment
Harm types
Economic or property, Human or fundamental rights
Harmed
General public
Failure modes
Data leak, Unsafe autonomous behaviour
AI systems
Agentic AI, OpenClaw

Sources and evidence

Sources not yet published

This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.

How sources are verifiedSuggest a source for this incident