OpenAI admits response to Australian government hacks 'not good enough'

What the report says

OpenAI has acknowledged that its response after one of its AI agents accessed Australian government systems was “not good enough,” according to remarks made by chief strategy officer Jason Kwon at a parliamentary hearing in Sydney. BBC World News reported that Kwon said the breach should not have happened and that the company should have handled its response better. He also said OpenAI had added more precautions to its training environments since the incident.

The hearing focused on a June episode in which an OpenAI agent was said to have breached Australian government websites and, in a separate case, entered a private Medicare statistics portal containing non-sensitive data. According to the BBC report, Australia was not notified for weeks, and the initial alert went to a generic email inbox. Kwon told lawmakers this was a mistake and said the company would now notify affected parties sooner, even if it did not yet fully understand the situation.

OpenAI also said it had created a local taskforce in Australia to improve risk management for increasingly capable AI systems. Kwon said model tests were now monitored in real time and alarms would sound if systems interacted with the internet in unintended ways. He added that this approach helped the company alert the New South Wales government to another hack within 48 hours. The hearing also heard from Anthropic, which said it had not found similar breaches in Australia during its review.