Speaking before an Australian parliamentary committee, Jason Kwon acknowledged that OpenAI should have notified Canberra sooner. Anthropic, for its part, supports a requirement to report serious incidents.
Jason Kwon, Director of the OpenAI Strategy, travelled to Australia to explain himself to the Joint Parliamentary Committee on AI in Sydney. This was his first public appearance since the Prime Minister revealed at the end of September that an agent of the company had entered a statistical portal in Medicare. He apologized and admitted that OpenAI should have alerted the government sooner rather than waiting for more evidence.
According to him, the company now warns its teams as soon as a model uses the Internet abnormally during its training. A second intrusion, targeting the New South Wales Parks and Wildlife Service, was detected last week and reported much faster. He also claimed that Sam Altman was unaware of the incident when he met Deputy Prime Minister Richard Marles on September 1, while employees had been aware for several weeks. Asked about Australians' particularly strong distrust of AI, he acknowledged that he could not explain it.
Anthropic and Copyright
Anthropic officials assured that they too would have made such an incident public, and supported the idea put forward by the Office of AI to compel developers to report serious security incidents. They reported finalizing an agreement for the Australian AI Security Institute to test their models. Their request to relax copyright to train models in Australia was, however, strongly challenged by the organizations of musicians, authors and dubbing actors.
Read the full article on the original source
Go to the source →