Singapore unveils new framework to rein in risks of autonomous AI agents

[SINGAPORE] Singapore has launched a new governance framework for agentic artificial intelligence (AI), as regulators move to set guardrails for the emerging technology which has far-reaching implications.

Unveiled during the World Economic Forum (WEF) in Davos, Switzerland, on Thursday (Jan 22), the Model AI Governance Framework for Agentic AI seeks to provide guidance to organisations on how to deploy agents responsibly, while making clear that accountability ultimately rests with humans.

It addresses the risks posed by agentic AI models, which are capable of independent reasoning, decision-making and executing multistep tasks with limited human oversight.

Agentic AI differs from traditional and generative AI, which typically respond to user prompts but lack the ability to initiate actions on their own.

Developed by the Infocomm Media Development Authority (IMDA), the framework builds on the 2020 Model AI Governance Framework. It follows consultation from both the public and private sectors.

The framework comes as organisations seek to advance their digital offerings and increasingly look towards agentic AI to drive their next stage of growth.

IMDA said AI agents would allow organisations to automate repetitive tasks and drive sectoral transformation by freeing employees’ time to undertake higher value activities.

While agentic AI is able to bring about positive changes to Singapore’s digital economy, the authority acknowledged that the agents come with vulnerabilities.

For instance, the use of AI agents with access to sensitive information, such as customer databases and financial records, might introduce new risks, including unauthorised and erroneous actions.

SEE ALSO

Asean agrees on guidelines for quicker subsea cable fixes

What’s keeping CEOs up at night? WEF poll points to insider threats, payment fraud, other cybersecurity issues

“It is therefore crucial to understand the risks agentic AI could pose and ensure that organisations implement the necessary governance measures to harness agentic AI responsibly,” said IMDA.

The guidance for deploying agentic AI responsibly spans four domains: assessing and bounding the risks upfront; ensuring humans are accountable; implementing technical controls; and enabling end-user responsibility.

The push for agentic AI is aligned with Singapore’s digital economy strategy, as the Republic seeks to expand its capabilities in the AI space while maintaining trust.

“Agentic AI is gathering momentum,” said Minister for Digital Development and Information Josephine Teo in Davos on Thursday, noting that there is interest in the business community to explore the usage of AI agents.

She added that as companies form opinions about the right way to implement agentic AI, the framework will help them make better and more informed decisions.

Asked about the liability of the agent executing an erroneous task, Teo acknowledged that it was “too early” to determine how specifically to assign liability for the agent itself.

However, she emphasised that regardless of how the service is provided, consumer protection laws still apply. Consequently, companies must assign internal roles and responsibilities to ensure accountability for AI agents.