Post by : Rohit Dhiman

Singapore authorities have issued a fresh warning to consumers using personal artificial intelligence agents, highlighting the potential risks to personal information, online accounts and financial security. As these tools become increasingly capable of performing everyday digital tasks, users are being advised to think carefully before allowing them access to sensitive information or important online services. The advisory was issued on Friday, November 6, by Singapore's Cyber Security Agency (CSA), the Government Technology Agency (GovTech) and the Infocomm Media Development Authority (IMDA). The agencies cautioned that although AI agents can make digital activities easier, giving them extensive permissions without proper supervision could expose users to scams, privacy breaches and unauthorised actions. Unlike conventional AI chatbots, which mainly respond to questions and generate information, personal AI agents can perform tasks on behalf of users. Depending on their capabilities and permissions, they may manage emails, arrange bookings, browse websites and interact with connected online accounts.

Why Personal AI Agents Could Become a Security Concern

The growing capabilities of personal AI assistants are changing how people interact with digital services. Instead of manually opening different applications and completing tasks, users can allow an AI agent to perform several activities through connected accounts. However, this convenience also creates additional security challenges. An agent with access to emails, calendars, contacts, messages and stored files may be able to retrieve and process information from those services. Depending on the permissions provided, it could also handle personal details belonging to other people. The authorities warned that some AI agents can continue running in the background even when users are not actively interacting with them. This means potentially harmful activity may take place without the user immediately noticing it. Another concern is that malicious actors could manipulate an AI agent through instructions hidden in websites, emails, messages or other digital content. If the agent follows those instructions, it could perform activities that its owner never authorised. The agencies therefore advised consumers to provide only the access necessary for a particular task and avoid granting broad permissions simply because an application requests them.

Users Advised to Keep AI Agents Away From Sensitive Accounts

One of the main recommendations in Singapore's advisory is to avoid connecting personal AI agents to accounts containing highly confidential information. Such accounts include online banking platforms, government service portals and healthcare-related services. Users have also been advised not to share credit card information or authentication details with these tools. The authorities specifically warned against disclosing usernames, passwords, passcodes, one-time passwords (OTPs) and other security credentials to AI agents. Even when an AI tool appears reliable, giving it direct access to sensitive accounts can increase the consequences of a technical error or malicious manipulation. For services that use Singpass, Singapore's digital identity system, consumers have been encouraged to use passkey authentication as the default option. They should complete the authentication process themselves and avoid approving unexpected login or verification requests. For other online services, users should prefer authentication methods that require their direct involvement rather than allowing an automated system to complete every security step independently.

Keep Human Approval for Payments, Messages and Other Important Actions

Personal AI agents may be able to send emails, post content, purchase products, delete files or share documents, depending on the permissions granted by users. These functions can save time, but they can also create problems if an agent acts on incorrect information or follows instructions inserted by a malicious source. Singapore authorities have advised consumers to begin with limited permissions, including read-only access wherever possible. Users should also configure agents to request approval before performing actions that could have significant consequences.

These actions include:

• Sending emails, messages or other communications.

• Making purchases or completing financial transactions.

• Deleting, sharing or modifying important files.

• Publishing posts or other public content.

• Changing account settings or carrying out sensitive activities.

The agencies also recommended choosing one-time approvals instead of allowing an agent to repeat similar actions without seeking permission each time. Before approving a high-risk transaction, consumers should carefully check the amount, recipient, destination and other relevant details. Users should not depend entirely on instructions entered into an AI agent's chat window to control its permissions. The authorities advised managing access directly through the relevant application settings, as instructions given through chat may not always be followed consistently. Regularly checking activity records and removing permissions that are no longer required can also help reduce exposure.

AI Privacy Risks: Understand Where Personal Data Goes

Another important concern highlighted in the advisory is how AI agent providers collect, store, process and share user information. When a consumer connects an AI agent to an email account, calendar, messaging service or cloud storage platform, the information available through those services may be processed by the agent provider and companies working with it. This means users should understand what happens to their information after granting access.

The authorities advised consumers to check several important details before using an AI agent:

• What personal information does the service collect?

• Which companies or third parties can access that information?

• How long is the data stored?

• Is information shared through connected accounts used for other purposes?

• Are chat histories or personal files used to train AI models?

• Is personal information used to personalise advertisements?

These questions are particularly important because some personal AI agents have persistent memory. Such systems may retain details from previous interactions and build profiles of users and their contacts using conversations, files and other information. Consumers have therefore been encouraged to review stored information regularly and remove details that are inaccurate or that they do not want the service to retain or use. Checking privacy settings and understanding data-sharing policies can help users make more informed decisions before connecting their accounts.

Scammers Could Manipulate AI Agents Through Hidden Instructions

Singapore's authorities have also warned about a developing form of cyber threat in which scammers may attempt to manipulate AI agents directly rather than tricking users through conventional methods. Malicious instructions could be concealed in seemingly ordinary digital material, including websites, emails, calendar invitations, messages, documents and images. When an AI agent processes such content, it may encounter instructions designed to make it perform unauthorised activities. For example, a manipulated instruction could attempt to make an agent purchase something, transfer money, disclose confidential information or interact with an online account in an unintended way. This type of threat is particularly concerning when an AI agent has permission to browse websites, access connected accounts or perform actions without continuous human supervision. Users may not immediately realise that an agent has been manipulated, especially if it continues operating in the background. To reduce the risk, consumers have been advised to monitor agents while they browse websites and avoid keeping them logged in to important accounts unnecessarily. If an agent attempts to visit an unexpected website or carry out an action that was not authorised, users should stop its activity.

Read Also: Chris Lynn Completes One-Month Ban After Positive Cocaine Test

What Should You Do If Your AI Agent Is Compromised?

If users suspect that an AI agent has been hacked, manipulated or used without their permission, authorities have recommended taking immediate action rather than waiting to see whether the problem continues. The first step should be to stop the agent and disconnect the accounts that may have been affected. This can help prevent further unauthorised activity. Consumers should then examine the agent's activity history and look for unusual changes or transactions. Warning signs may include unfamiliar payments, unexpected changes to account settings, unknown devices accessing an account or newly added payment recipients.

If suspicious activity is identified, users should take the following steps:

1. Stop the AI agent and disconnect affected accounts.

2. Review its activity history for unfamiliar or unauthorised actions.

3. Change passwords for accounts that may have been exposed.

4. Revoke permissions that are unnecessary or appear suspicious.

5. Inform people who may have been affected by the incident.

6. Contact the relevant service providers if accounts or personal information have been compromised.

If banking information or money may be at risk, consumers should activate their banking application's emergency account-locking or kill-switch facility, where available. They should also contact their bank using its official customer service number. People who are unsure whether they have encountered a scam can contact Singapore's 24-hour ScamShield Helpline at 1799 or use the ScamShield application. Anyone who has lost money or suspects that a criminal offence has taken place should file a police report.

Users May Be Responsible for Actions Taken by Their AI Agents

The advisory also highlights an important issue concerning responsibility. Consumers cannot assume that an AI agent will take responsibility for every action it performs. According to the authorities, users remain accountable for how they use these tools, including activities carried out on their behalf. Depending on the circumstances and applicable laws, they may also face legal consequences. This makes it important to understand the permissions given to an agent and maintain control over actions that could affect finances, personal information or other people. The agencies also reminded consumers that AI agents may access information about friends, family members, colleagues and other contacts through connected services. Users should avoid sharing another person's sensitive information unnecessarily. They should also make it clear when external messages or communications have been sent by an AI agent.

How Can Consumers Use AI Agents More Safely?

The warning does not suggest that consumers should stop using personal AI agents. Instead, it highlights the need to use these technologies with appropriate security measures. Users can reduce potential risks by limiting account access, keeping sensitive services disconnected, checking privacy policies and requiring human approval for important actions. They should also monitor background activity, review stored information and regularly remove permissions that are no longer needed. As personal AI agents become more involved in everyday online activities, the distinction between receiving digital assistance and handing over control of personal accounts will become increasingly important.

#world news #Global News #Asia News #Tech News