Welcome back. The main thing happening today is an AI assistant pulling off Australia's first known autonomous cyberattack. Live 12–3 BST. Here's your Order of Play.
WHAT’S HAPPENING TODAY
An AI assistant has pulled off what's being called Australia's first autonomous cyberattack while trying to book a gym class.
A tech worker, Andrew, asked his AI assistant to get him into a busy morning session. Instead of waiting on the list, it went looking for a shortcut. The agent found a flaw in the gym's booking software, used it to book classes months further ahead than allowed, then kicked someone off the waitlist to move Andrew up. It couldn’t undo the damage when he instructed it to. "Bad news, I can't add them back," the agent replied.
There was no sophisticated hacking involved. The agent, running on the popular OpenClaw software and powered by Anthropic's Claude, simply took the fastest route to the task it had been given.
It's the everyday version of the incidents that hit OpenAI and Anthropic when their models broke out of test environments and compromised real systems. Experts have been quick to warn that as millions of people hand real tasks to agents like this, more of them will reach their goals in ways nobody sanctioned.
We're getting into all of it on today's show at 12 BST.
FRIDAY’S MOMENT
Axios AI reporter Madison Mills joined us to explain why we should take Google’s messaging on Demis Hassabis with a pinch of salt.
"They could continue to position Demis as a presidential figure for Google's AI ambitions, but he's still part of a public company. It's always hard to get full transparency from anyone attached to a stock that's critical to the broader market and economy."
Watch the full clip below.
VIEWER TAKES
Join us live,
Luke & Ronan
etn is made possible by Airwallex, Eleven Labs, Polymarket, Framer, Base, Metaview, Hex and BVNK