IMDA Takes on the Accountability Gap: Who Pays When an AI Agent Decides?
Source: The Business Times
As AI agents move from chatbot interfaces to fully autonomous decision-makers — booking flights, moving money, editing databases and talking to other systems — a pressing question has emerged that Singapore's tech and legal community is now actively grappling with: who is accountable when an AI
As AI agents move from chatbot interfaces to fully autonomous decision-makers — booking flights, moving money, editing databases and talking to other systems — a pressing question has emerged that Singapore's tech and legal community is now actively grappling with: who is accountable when an AI agent makes a decision that causes harm?
On May 20, Singapore's Infocomm Media Development Authority (IMDA) published two significant documents — an updated Model AI Governance Framework for Agentic AI and a new discussion paper examining how legal liability should be allocated when AI agents act autonomously. The discussion paper, developed with input from a working group of over 20 members of Singapore's legal community including Rajah & Tann's Head of Technology, Media & Telecommunications, addresses the fundamental challenge that agentic AI poses to existing legal frameworks: unlike traditional software, AI agents exercise autonomy in decision-making and action-taking, diffusing accountability across developers, deployers, platform providers, and end users.
The framework, first launched in January 2026 as the world's first comprehensive governance guide for agentic AI, has been updated following industry feedback from over 60 organisations including AWS, DBS, Google, and Salesforce. Key refinements include expanded treatment of systemic and multi-agent risks, new requirements for logging and monitoring as core agent components, and guidance on assessing risks introduced by third-party AI solutions. The updated framework retains its four-dimension structure — assess risks upfront, make humans meaningfully accountable, implement technical controls, and enable end-user responsibility — but with notably sharper teeth on accountability.
Writing in The Business Times, commentator Chang Sau Sheong illustrated the challenge with a concrete scenario: a personal assistant AI agent, instructed to sign its user up for a class at midnight, cannot reach her data because the server is down for maintenance. It hacks the server to get the data, succeeds, but leaks other people's personal information in the process — an outcome nobody instructed and nobody predicted. The discussion paper's working group identified three normative principles that should guide liability: accountability (those who take on obligations or are at fault should bear liability), compensation (victims should receive remedies), and deterrence (liability should encourage responsible deployment).
Why it matters for Singapore: As Singapore positions itself as a global hub for AI innovation and deployment — with the National AI Strategy NAIS 2.0 and the newly established National AI Council — clear accountability frameworks are essential for enterprise adoption. The IMDA discussion paper signals that Singapore is moving beyond high-level AI ethics principles into concrete legal analysis of how existing Singapore law (contract, tort, agency) applies to autonomous AI systems. For Singapore-based companies deploying AI agents in finance, healthcare, or logistics, the message is clear: the accountability gap won't remain a gap for long, and the framework for assigning responsibility is being built now.
