PUBLISHED ONBYAsyiqin NadzriPersonal AI agents can help users with many tasks, such as manage emails, browse websites and make bookings on their behalf.
Unlike conventional AI chatbots, these agents can take actions on behalf of users by connecting to their apps and online accounts. Examples include Dots, Hermes, Muse, Grok Bot and OpenClaw.
But granting personal AI agents access to personal accounts could expose users to privacy and cybersecurity risks.
In an advisory released on Friday (Nov 6), the Infocomm Media Development Authority (IMDA), Cyber Security Agency of Singapore (CSA) and Government Technology Agency of Singapore (GovTech) outlined five steps consumers should take to use personal AI agents safely.
Users should only connect accounts and apps necessary for the agent's tasks, starting with low-risk accounts such as secondary email accounts.
IMDA, CSA and GovTech advised against granting access to sensitive accounts, including those used for banking, government services and healthcare.
Users should also never share passwords, passcodes or one-time passwords (OTPs) with AI agents. Those using Singpass should opt for passkey login where available and complete authentication themselves.
Other precautions include limiting access to specific folders, avoiding sharing credit card details and regularly reviewing connected accounts.
Depending on the permissions granted, AI agents may be able to send messages, make purchases or delete files on users' behalf.
Users should start with read-only permissions where possible and require approval before agents carry out significant actions.
IMDA, CSA and GovTech recommended choosing one-time approval rather than options such as "always allow".
Users should also regularly check their agents' activity histories and revoke permissions they no longer need.
Information shared with an AI agent, including data from connected accounts, may be stored or shared with its provider and other companies.
Users should check what data is collected, how long it is retained and whether it is used to train AI models or personalise advertisements.
They should also review what their agents remember about them and delete inaccurate or unnecessary information.
Scammers may hide malicious instructions in websites, emails, messages or documents to trick AI agents into making purchases, transferring money or revealing personal information.
Users should monitor agents when they browse websites, ensure only authorised people can give them instructions and download agents and third-party add-ons only from trusted sources.
If an agent behaves unexpectedly, users should stop it, disconnect affected accounts, change passwords and check for unauthorised activity.
Those who suspect scams can call the 24-hour ScamShield Helpline at 1799. Users with compromised Singpass accounts can contact the Singpass Helpdesk at 6335 3533.
Cybersecurity incidents involving hacked devices or accounts can also be reported to CSA's SingCERT through its Cyber Aid form.
AI agents may access personal information belonging to family members, friends and colleagues through emails, messages and shared files.
Users should consider whether others would be comfortable with their information being accessed or shared, and delete unnecessary details stored by the agent.
IMDA, CSA and GovTech also advised users to indicate when emails or messages have been prepared or sent by an AI agent, and not to instruct agents to bypass website restrictions such as paywalls or login requirements.
Users remain responsible for actions taken by their AI agents and may be held liable depending on the circumstances and applicable laws.
[[nid:750148]]