Singapore authorities urge strict limits on AI agent autonomy, access to personal accounts
Personal AI agents can take actions on users’ behalf, such as browsing websites, managing emails and making bookings.
The Cyber Security Agency of Singapore, in collaboration with the Government Technology Agency and Infocomm Media Development Authority, has released an advisory on Nov 6 urging individuals to employ strict limitations when managing personal artificial intelligence agents. These agents, which can execute actions on behalf of users, such as browsing websites, managing emails and making bookings, pose both convenience and privacy risks.
The advisory warns that agents may expose sensitive information, make errors, or exceed users' intentions if granted broad access to personal accounts and applications. To mitigate these risks, the advisory provides five precautionary steps for safe usage. Firstly, users should only connect the necessary apps and accounts to agents and, when feasible, restrict access to specific information within them.
Initially, users should limit access to less sensitive material, such as a secondary email account, to familiarize themselves with the agent's functioning. Connecting to sensitive accounts, such as banking, government, or healthcare accounts, is strongly discouraged. Users should refrain from sharing credit card details or authentication credentials, including passwords and one-time passwords.
The second precaution involves retaining control over crucial actions, starting with limited or read-only permissions where feasible. Consent should be sought before actions with significant consequences, such as sending messages, deleting or sharing files, or posting publicly. Users should avoid "always-allow" options and scrutinize details meticulously for high-risk transactions.
Lastly, users must understand how their information is processed and stored by the agent provider and its partners. It is crucial to review what data is collected, who receives it, the retention period, and the post-disconnection circumstances. Users should also monitor whether their data is utilized for AI model training or personalized advertisements and adjust settings accordingly.
The advisory cautions against trusting agent instructions blindly, as they may not always be followed consistently. Instead, users should manage permissions through app settings and regularly revoke unnecessary permissions. Additionally, users must remain vigilant against scams and malicious actors who may target agents through deceptive instructions hidden in various digital formats.
The best defense is to limit an agent's visibility or functionality, restrict instruction-givers, employ robust authentication methods, and download agents solely from official sources.
Written by urgent.news from Straits Times's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
