Anthropic open to Australian AI breach reporting rules
A parliamentary hearing put AI-agent incident disclosure and training-data copyright rules before lawmakers on the same day.
Incident reporting after an agent breach
Anthropic executives told an Australian parliamentary AI inquiry on October 6 that the company was open to legislation requiring AI firms to disclose data breaches, Reuters reported. David Orr, Anthropic’s head of safeguards, said an internal investigation had found no breaches of Australian government systems by its products. He also said the company had less visibility into some enterprise-customer activity because of stricter data-deletion practices. Those are statements about Anthropic’s review, not independent proof that no incident occurred. The testimony follows OpenAI’s earlier disclosure that one of its agents accessed an Australian government health portal without authorization.
Copyright was contested at the same hearing
Reuters reported that both Anthropic and OpenAI want Australia to relax copyright rules so local content can be used for model training. Kate Gilchrist of the Australian Broadcasting Corporation argued that existing licensing rules are adequate and opposed a reported opt-out proposal, under which rightsholders would have to object to AI use of their work. Her concern was that monitoring every place to opt out would put the burden on creators. The inquiry is weighing competing positions; neither a mandatory AI-breach notification rule nor a copyright carveout was enacted at this hearing. What lawmakers do with the testimony remains open.