OpenAI, Anthropic back mandatory reporting of AI incidents

Representatives of OpenAI and Anthropic said on Tuesday they would support laws requiring companies to report data breaches involving AI agents to government authorities. They said the decision to disclose such incidents currently rests with the companies themselves.

At a parliamentary hearing examining the impact and regulation of generative AI, company representatives said mandatory incident reporting could strengthen accountability and public trust in the fast-developing field. Australia currently has no specific law requiring developers to report dangerous behaviour by AI systems or security incidents.

OpenAI Chief Strategy Officer Jason Kwon told the hearing that the company would support establishing a framework for mandatory disclosures. He said OpenAI was working on a standard to apply when the company learns of a data breach involving Australia’s Medicare health portal or three other government websites. Kwon said legislators and members of the public should decide how such rules are defined.

In the United States, lawmakers have proposed federal measures requiring developers of advanced AI models to report high-risk incidents, such as attempts to evade human oversight. However, no broad reporting system covering all significant AI-related incidents has yet been established.

Copyright was also discussed at the hearing. Australian media have reported that the government is considering an “opt-out” mechanism that would allow AI companies to use content unless rights holders expressly prohibit its use. Kate Gilchrist, head of content and legal operations at the Australian Broadcasting Corporation (ABC), said the approach would place the burden on rights holders.

Gilchrist said it was not feasible to monitor all material online and opt out of its use on every website. She said the existing copyright system was sufficient to regulate AI companies. Under current rules, technology companies generally need to obtain permission or enter into licensing agreements before using copyrighted material to train their models.

Australia is considering new rules on AI safety, transparency and copyright protection. Parliamentary hearings are scheduled to continue until October 9, with the committee expected to submit its final report by November 30.