Singapore Government Releases AI Agent Sandbox White Paper on Public-Service Benefits and Governance Risks

AI agents can independently break down tasks and operate websites. In public services, they could reduce the burden of manual testing and application processing, but also amplify the risks of errors, personal-data leaks and unauthorized actions. Singapore's Cyber Security Agency, Government Technology Agency and Infocomm Media Development Authority therefore partnered with Google to test the benefits and governance requirements in a controlled sandbox. The white paper did not disclose the value of the partnership.

The three agencies released the white paper on May 20, 2026. Launched in August 2025, the sandbox ran for about four months and tested government-website quality assurance, chatbot security assessments and social-welfare applications. The agents identified broken pages, placeholder text and misplaced test URLs, and organized simulated application data. The report also warned that indirect prompt injection could trigger unintended actions, calling for layered defenses and human review calibrated to risk.