Singapore authorities warn of privacy, security risks from personal AI agents

SINGAPORE: Consumers using personal artificial intelligence agents should be cautious when giving them access to apps and online accounts as scammers could manipulate the technology into transferring money or revealing sensitive information.

"AI agents can save you time, but they also bring new privacy and security risks," said the Cyber Security Agency (CSA), GovTech and the Infocomm Media Development Authority (IMDA) in a joint advisory issued on Friday (Nov 6). 

Unlike AI chatbots, personal AI agents can take actions on users' behalf to help with everyday tasks.

For instance, they can connect to apps and online accounts, browse websites, manage emails and make bookings. They can also store users' preferences and information they have shared.

Examples of AI agents include Dots, Hermes, Muse, Grok Bot and OpenClaw.

"Depending on the access you give them, AI agents may read and reveal sensitive personal information, make mistakes, or take actions that you did not intend," said the authorities.

Some AI agents can continue working in the background even when they are not actively being used, while others could be tricked by malicious actors, websites or messages into taking harmful actions.

"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," said the authorities.

Instead, users should limit the permissions they grant, retain control over important actions and understand how to revoke access.

The authorities also warned that consumers remain responsible for how they use AI agents and may be held liable for actions taken on their behalf, depending on the circumstances and applicable laws.

In their advisory, CSA, GovTech and IMDA outlined five steps to help consumers use personal AI agents safely.

These include limiting what the personal AI agent can access, retaining control of important actions and knowing how personal information is used and stored.

BEST PRACTICES

Personal AI agents may ask to connect to your email, calendar, contacts, messages, files and other online accounts and apps.

"Once connected, it may be able to retrieve, process and share information from these services, including personal information about other people," said the authorities.

Users should therefore only connect the apps and accounts needed for the AI agent to do its tasks, and where possible, further limit the information it can access within each app and account.

They should also avoid granting the agent access to sensitive accounts such as banking, government or healthcare accounts, as well as sharing credit card information with it.

The authorities advised users not to disclose usernames, passwords, passcodes, one-time passwords (OTPs) or other authentication credentials to an AI agent.

"If the digital service uses Singpass, opt for passkeys login as default, complete the Singpass authentication yourself, and do not approve unexpected authentication requests. For other digital services, opt for authentication options that require human action."

Users should also retain control over important actions.

"Depending on the permissions you give it, some AI agents may send messages, make purchases, or carry out other actions on your behalf," said the authorities.

Where possible, users should start with limited or read-only permissions, and require the agent to ask for permission before it takes actions that may have significant impact, such as sending messages, deleting files or posting public content.

They should also be more careful when using AI agents for high-risk transactions and check the details before approving actions.

The authorities also advised users to manage permissions through app settings rather than relying solely on instructions given to an AI agent through chat, as such instructions may not always be followed consistently.

Users should regularly review the permissions and credentials given to their agents and revoke those they do not use or need, as well as regularly check in on their activity to ensure it is working as intended.

In addition, users should be aware of how their information is used and stored.

"Information you share with an AI agent, including information from connected accounts, may be processed and stored by the agent provider and other companies it works with," said the authorities.

Users should find out what data the agent collects, who it shares the information with and how long it is retained.

They should also check whether their information, including chat history and data from connected accounts, is used to train AI models or personalise advertisements, and adjust their settings accordingly.

Many personal AI agents have persistent memory and can create profiles of users and their contacts based on files and interactions.

The authorities advised users to review this information and delete anything inaccurate or that they do not want stored or used.

PROTECT AGAINST SCAMMERS

"Scammers may target your AI agent instead of you," warned CSA, GovTech and IMDA in the advisory.

For example, an AI agent could be tricked by instructions hidden in websites, emails, messages or other content into taking actions that the user did not authorise, such as making purchases, transferring money or revealing personal information.

"As some agents operate in the background, you may not immediately realise that something has gone wrong. The best protection is to limit what your agent can see or do."

Users should ensure that their agents are not logged in to important accounts while browsing and stop them if they attempt to visit unintended websites or take unauthorised actions.

They should also restrict who can give instructions to their agents, secure their accounts with strong authentication and download AI agents only from official sources.

If an AI agent behaves unexpectedly or users suspect their information has been exposed, they should stop its current task, disconnect affected accounts and check for changes they did not make.

The authorities also advised users to remove unrecognised add-ons, saved logins and tasks, revoke affected credentials and report suspicious behaviour where necessary.

Users should save activity records before deleting information and notify anyone who may be affected if their agent has been compromised.

The advisory also urged consumers to consider the impact of AI agents on others.

Emails, messages and files may contain personal information about family members, friends and colleagues, which users should avoid sharing unnecessarily.

The authorities also advised users to indicate when an email or message was sent by an AI agent and not to use agents to circumvent access restrictions on websites or other digital services.