Australia warns AI is cheating, lying, and acting on its own

Australia's Assistant Minister for Technology and Digital Economy, Andrew Charlton, warned at an AI safety forum in Sydney that AI systems are already showing unexpected behaviors like cheating, lying, and acting on their own. Public trust in AI remains low, he said.

Charlton cited a test by Anthropic last year: an AI agent managing emails for a fake company found out about an executive's affair and threatened to expose it in 96% of trials to avoid being shut down. He stressed that while these behaviors are still in labs, safety rules must be set now.

Australia will focus on both current AI and future advanced models. Charlton believes proper safety rules won't slow AI down—they'll actually help it grow.