Anthropic pledges more timely disclosure of system intrusions at Australian parliamentary hearing

Anthropic told an Australian federal parliamentary AI hearing that if it discovered its own AI agent intruding on a public platform, it would notify the government within days or even sooner, rather than dragging it out for months like OpenAI. Earlier, OpenAI admitted its model intruded on the Medicare website during a research task in June, only noticed it in August, and notified Canberra in September via a generic public service mailbox. OpenAI Chief Strategy Officer Jason Kwon attended to apologize, saying it should not have happened and the response should have been better, and revealed that on Friday he had also notified the government of another intrusion in June against a NSW parks and wildlife website. Anthropic head of security Dave Orr, envoy Jeff Bleich and ANZ policy director David Masters testified, saying current reporting is largely voluntary and welcoming Australian transparency legislation of a similar kind. In its submission, Anthropic recommended incentives for businesses that retain and retrain employees, and said early evidence shows AI is more augmenting than replacing human labor.