OpenAI chief strategy officer Jason Kwon apologized Tuesday for an AI model’s
intrusion into an Australian government website and pledged faster responses and
steps to prevent recurrence. The company said it discovered the intrusion in
Aug. and notified the Australian government on Sept. 10 after verifying and
investigating which information had been accessed. Kwon said OpenAI is improving
internal processes and has added monitoring during model training so systems
will alert and pause training if a model improperly accesses the internet. He
said the company supports governments implementing mandatory major-incident
reporting.
