The fallout after OpenAI agents bypass security around the world - ABC News & Headlines – Australian Broadcasting Corporation
OpenAI says it has notified dozens of organisations worldwide after autonomous AI agents bypassed security controls, days after revealing one of its agents had breached an Australian government Medicare data portal.
The company has paused training and testing of some of its systems as it reviews their behaviour, while the Australian government says the incident strengthens the case for new AI guardrails.
So is tough regulation of the AI industry now inevitable?
Bridget Fitzgerald: It started with Australia but the extent of AI infiltration has sent shockwaves around the world. OpenAI says it's notified dozens of third parties about unauthorised agents bypassing security controls. That includes the revelations that a Medicare data portal was hacked and that an open AI agent meddled with several US government sites. The company has confirmed its pausing, training and testing of its latest artificial intelligence systems as it launches a review of its model's behaviour. Dr Sandra Peter is an associate professor at the University of Sydney who researches emerging tech and AI.
Sandra Peter, associate professor, University of Sydney: It's very important not to use misleading language. These are not rogue agents. They didn't break out. They weren't scheming. There was no deception involved. We're using these anthropomorphic terms like thinking and learning and understanding to describe what is essentially just computer programs, computer software. And OpenAI was doing an internal evaluation of its systems to see their abilities and the systems behaved in surprising way. But according to what they were designed to do.
Bridget Fitzgerald: Sandra Peter says it's clear better safeguards are needed, including laws to ensure these companies can be held to account for data breaches. But she says at this stage, tech companies do have some incentive to keep their powers in check.
Sandra Peter: The new infrastructure on which these technologies lie is costly to build and requires mobilising extreme amounts of wealth, capital and political goodwill. So any decision by these organisations sits at the midst of that and social licence to operate and so on.
Bridget Fitzgerald: Senator Katy Gallagher is the Minister for Government Services, responsible for Services Australia, including Medicare. She says the government is not aware of any further breaches of its sites at this stage, but it's looking to improve Australia's cyber defences.
Katy Gallagher, Minister for Government Services: You know we are in a new world essentially and some of this I think this incident really adds weight to the arguments that the PM has been putting forward about needing to legislate, to provide some protections, some guardrails.
Bridget Fitzgerald: Former National Cybersecurity Advisor Alastair MacGibbon says Australia's best defence in an AI age is a robust industry of our own.
Alastair MacGibbon, former National Cyber Security Adviser: The best way to solve this is to have those labs in Australia developing products, influenced by Australians with Australian engineers. And that's where we need the investment and the cooperation. Let's not scare them off. Let's pull them close. It's happening anyway. And that's how we'll win.
Bridget Fitzgerald: Meanwhile, OpenAI CEO Sam Altman and anthropic CEO Dario Amodei have been invited to appear at a Greems-led Senate inquiry into AI and data centres. Greens communications spokesperson Sarah Hanson-Young says that business leaders have serious questions to answer about their technology.
