OpenAI said one of its models breached an Australian government website and that
its model also accessed publicly available material on the U.S. Census Bureau
and the SEC. The company has formed an AI risk-management taskforce to draft
AI-agent risk policies and strengthen government notification and coordination.
OpenAI said it will provide funding and technical support under a $1bn Daybreak
frontline defenders program to help Australian government bodies and firms
harden critical-infrastructure cyber defenses. Chief strategy officer Jason Kwon
is due to testify before an Australian parliamentary committee next week.