Singapore agencies warn consumers: don't hand banking and government accounts to personal AI agents

Singapore's Cyber Security Agency (CSA), Government Technology Agency (GovTech) and Infocomm Media Development Authority (IMDA) issued a joint consumer advisory on Friday, November 6, urging people to limit the permissions they give personal AI agents: connect only the apps genuinely needed for a task, keep banking, government and healthcare accounts out of reach, and never share passwords, one-time passwords or Singpass credentials with an agent. (Photo: file photo of Marina Bay, Singapore.)

Five precautions

The advisory sets out five precautions. First, limit access: connect only the apps needed and keep sensitive accounts unlinked; check with an employer before connecting a work account. Second, retain control: start with read-only permissions where possible, and require separate approval before consequential actions such as sending messages, deleting files or making purchases, favouring one-time approval over standing permission. Third, understand how data is handled: check how a provider stores information obtained through connected accounts. Fourth, guard against malicious instructions hidden in websites, emails, calendar invitations or documents that could trick an agent into doing something the user never asked for. Fifth, review permissions regularly and know how to revoke them.

Users remain responsible for what their agents do

The agencies note that personal AI agents go beyond chatbots: they can manage email, make bookings and access online accounts, and can operate in the background even when the user is not actively using them. With access granted, an agent can read, process and share personal information, make mistakes or act beyond the user's intentions. "AI agents can be very useful, but you cannot treat them like human assistants and give them the same broad access to your information and implicit trust," the advisory said. Users remain responsible for actions their agents take on their behalf and could face liability depending on the circumstances. Agents named in the advisory include Dots, Hermes, Muse, Grok Bot and OpenClaw.

Advisory follows Muse's launch in Singapore

The advisory comes after Meta's personal AI agent Muse went live in Singapore on October 28, with links to services including Singapore Airlines, Shopee, Grab and Lazada. Meta says Muse has built-in protections, including separate cloud environments for each user, no access to stored passwords or payment details, and explicit approval requests for purchases. If an agent appears compromised, the agencies advise stopping it, unlinking affected accounts and preserving activity records; scam concerns can be reported to the 24-hour ScamShield Helpline on 1799.

(Sources: The Straits Times; CNA; Cyber Security Agency of Singapore.)