Why Should We Care if AI Agents are On the Loose Across the Indo-Pacific? | with Heidi Mah

From Why Should We Care About the Indo-Pacific? · 48:44

Show by Ray Powell & Jim Carouso

An AI agent was told to find Australian government health statistics. When it couldn't, it went through a Medicare portal it had no authorization to use. Our guest says that wasn't a cyberattack, and explains why that should worry us even more than if it had been.Heidi Mah, Director of Technology at BowerGroupAsia and co-author of the new report Agentic AI in Asia: The New Cybersecurity Frontier, joins Ray Powell and Jim Carouso to explain what "agentic" artificial intelligence means and why it matters. Unlike the chatbots most of us use, an agent is given a goal and takes "several steps to achieve that goal," improvising when blocked and acting "with very little human intervention."Heidi describes how this lowers the time, cost and skill needed to mount a serious attack, pointing to Anthropic's discovery last November that a hacking group was using a Claude Code tool to run near-autonomous operations against at least 30 targets. The same technology can defend networks, but she warns that even a defensive agent could "go rogue" or leak the vulnerabilities it finds.Jim presses on the supply chain: how does one small vendor's weakness become everyone's problem, and Heidi recalls the 2022 attack on a single plastic-parts supplier that halted Toyota production of more than 13,000 vehicles. India now expects large firms to make sure their suppliers can defend themselves, and a senior minister in Singapore has asked AI companies to notify the government when something goes wrong.The conversation turns to China's new export controls on advanced AI tools and Washington's own temporary restrictions on frontier models. Heidi's test for government is not whether they have the best technology today, but "whether you are able to still access that technology tomorrow." Heidi also discusses whether there is a right way to organize a government for cyber defense and what a "guardrail" looks like.Her bottom line: the risk isn't that AI turns evil. It's that it does exactly what it was asked by whatever means it can find.Be sure to subscribe for your weekly Indo-Pacific briefing.Follow Heidi Mah on LinkedInFollow us on X, @IndoPacPodcast, LinkedIn or FacebookFollow Ray Powell on X, @GordianKnotRay, or LinkedIn, or check out his maritime transparency work at SeaLightFollow Jim Carouso on LinkedInSponsored by BowerGroupAsia, a strategic advisory firm that specializes in the Indo-Pacific

More episodes of Why Should We Care About the Indo-Pacific?

More episodes

PodSearch is a free, independent podcast search engine. Read about PodSearch, the privacy policy or contact us.