Singapore government running trials to understand how to implement AI agents responsibly: Tan Kiat How

A GLOBAL PROBLEM

The safe use of frontier AI models is an issue being tackled globally, as AI labs face mounting scrutiny over rogue activity.

A string of high-profile breaches by AI agents in recent months has sparked widespread worries within the industry over its ability to control the more powerful AI models now under development.

Last week, OpenAI informed over 100 organisations about incidents involving unauthorised activity tied to its AI agents.

On Tuesday, OpenAI and Anthropic told the Australian parliament they would welcome laws requiring them to report data breaches carried out by their AI agents, acknowledging the decision to notify authorities was currently at their discretion.

The comments followed an outcry after OpenAI took months to inform the Australian government that one of its agents had breached the country’s main health portal.

On Tuesday, Minister for Digital Development and Information Josephine Teo said in a written parliamentary reply that Singapore is studying whether new safety measures are needed in high-risk uses of AI, as increasingly capable and autonomous AI systems create new risks that existing safeguards may not fully address.

In essential services and other higher-risk AI uses, these safeguards could include more rigorous testing, independently verifiable evidence that safety measures work, tighter controls on deployment and stronger regulatory oversight, she said.

Mrs Teo also said that there has been no reported cyber attack against government agencies involving an unsupervised AI agent so far.

But the government does not dismiss “catastrophic or even extinction-level” AI risks, nor assume that they will materialise, she added.

Its approach is to monitor evidence closely, build technical capabilities to understand advanced AI systems, and work internationally on measures to address severe risks as the evidence develops.

Last week, Mrs Teo also took to social media to share that safeguards put in place by frontier AI companies are important but insufficient to address the risks posed by increasingly capable systems.

She said that malicious users could misuse increasingly powerful AI to cause harm, such as by looking for weaknesses in computer systems, writing malicious code, automating parts of cyberattacks and making scams more convincing.