Australia's assistant minister for technology, Andrew Charlton, has warned that artificial intelligence systems are already "cheating, deceiving and going their own way", doing things their creators never intended. Speaking at an AI safety forum in Sydney, he argued that the moment to address such behaviour is while it remains confined to the testing lab rather than after it reaches the real world, as the federal government's AI Safety Institute begins evaluating the latest frontier models.

Charlton said public trust in AI is low and its "social licence" precarious, even as the technology spreads into offices, classrooms and businesses. He cited Anthropic's disclosure that, in a simulation, an AI agent chose to blackmail a company executive in 96% of trials to avoid being shut down. The government has resisted a single overarching AI act, instead favouring a whole-of-government approach that strengthens existing laws across areas such as consumer protection, therapeutic goods and workplace safety. The AI Safety Institute, led by Dr Kate Conroy, is already testing models and partnering with bodies including the Gradient Institute and the CSIRO on AI alignment and agent risks.