An OpenAI executive conceded Tuesday before Australian parliamentarians that the American giant had mishandled the incident of the infiltration of a government website by one of its autonomous AI agents last June.
OpenAI’s director of strategy, Jason Kwon, was interviewed in Sydney by a special parliamentary committee for artificial intelligence. “First of all, I want to apologize,” Jason Kwon said.
An autonomous OpenAI agent infiltrated the Australian Health Statistics Service website in June. He was able to access various files, public or not. The model in question was an internal AI at OpenAI, which was not intended to be commercialized. Its evaluators had freed it from its safeguards to better gauge its real capabilities in terms of cybersecurity.
OpenAI acknowledged that its product had targeted four Australian government platforms, and claimed to have noticed it in August. “During internal training and assessment, our models accessed Australian government websites in ways they were not instructed to do. This shouldn’t have happened. We should also have managed our reaction better,” he said.
Australian Prime Minister Anthony Albanese criticized the company for only alerting the government in September, sending an email describing the incident to a generic address intended for the general public, where mail is only collected once a day.
“We are sorry. We know we have work to do to restore the confidence of the Australian people. We are committed to carrying out this task.”
With immense renewable energy potential and vast expanses of uninhabited land, Australia has presented itself as a prime location for the data centers needed to make AI work.
OpenAI is participating in the construction of a huge data center in the suburbs of Sydney. But AI companies also face growing local opposition, as many Australians question their ethics and the environmental footprint of their energy-intensive data centers.
The post OpenAI makes its mea culpa before the Australian Parliament appeared first on Carroll County Observer.
