CSA Advisory on Personal AI Agents Tells Singapore Users to Keep Them Away From Bank and Singpass Logins

Singapore's Cyber Security Agency, GovTech and IMDA issued a joint advisory on personal AI agents on 6 November, warning that assistants able to act inside a user's apps and accounts can be turned against them. Its eight-page guidance names five products: Dots, Hermes, Muse, Grok Bot and OpenClaw. The Straits Times notes it follows the Singapore launch of Meta's Muse.

The risk: scammers instructing the agent

The agencies' central worry is hidden instructions. Scammers may "target your AI agent instead of you", planting commands in web pages, emails or calendar invitations that push it into buying something or moving money. Add-ons are the second route in: the guidance says malicious ones "could steal your information" and warns of look-alike apps and fake installers. Some agents keep running in the background and build profiles of their owners and contacts. The user may also "be held liable for these actions", the agencies add.

What the agencies tell users to do

Don't connect banking, government or healthcare accounts at all. Never hand the agent a password, passcode, OTP or credit card number. Singpass users should make passkey login their default, and a separate email address can receive OTPs out of the agent's sight.

For permissions, the advice is read-only access first and "one-time approval over 'always-allow'". Telling the agent in chat what not to do isn't enough, the PDF says; limits belong in its settings. Agents should be downloaded only from official sources and kept out of group chats where strangers can send them instructions.

An annex walks through the settings of Muse 14.1 on iOS.

If an agent misbehaves

Stop it, disconnect its accounts, change passwords and look for changes made without approval. Suspected scams go to the ScamShield Helpline on 1799; Singpass problems to its helpdesk on 6335 3533; incidents to SingCERT through CSA's Cyber Aid form.