Skip to main content

Singapore warns of account and payment risks from personal AI agents

Authorities say malicious content can trick agents into unauthorised actions, and recommend keeping sensitive accounts beyond their reach.

By Arif Rahman

Illustration of an AI assistant interface beside account security controls
Photo: CNA

Singapore’s Cyber Security Agency, GovTech and Infocomm Media Development Authority issued an advisory on Nov 6 cautioning consumers that scammers could exploit personal AI agents to move money or expose private data. They urged users to keep account access narrow and retain oversight of consequential decisions.

Personal agents go beyond chatbot conversations by performing tasks for users, from handling email to arranging bookings. The agencies warned that permissions can also enable them to disclose confidential material or act against a user’s intentions.

Attackers could plant instructions in content encountered by an agent, such as a document, image or calendar invitation. That could trigger unauthorised spending or disclosure. Agents running without active supervision may also leave users unaware of a problem.

The agencies recommended excluding banking, healthcare and government accounts, and withholding payment card details and authentication credentials. For Singpass-enabled services, they advised choosing passkeys as the default and personally completing authentication.

They also recommended beginning with read-only or otherwise restricted access, requiring separate permission for consequential actions, and controlling permissions through application settings. Chat-based directions alone may not be followed reliably. Users could face liability for an agent’s actions, subject to the circumstances and relevant laws.

Consumers should examine providers’ data practices, including retention, sharing, model training and advertising uses, the agencies said. They also advised checking persistent memories and removing inaccurate or unwanted material.

If an agent appears compromised, authorities recommended halting it, unlinking affected accounts and inspecting its activity history. Where money could be at risk, users should activate their bank’s kill switch or lock payment cards, then call the bank’s official hotline. Suspected crimes or financial losses should be reported to police.

Keep reading: AI

All AI stories

Latest in Technology

All latest