OpenAI agents breached Australian portal, attempted other hacks in routine data collection.
Australian PM said OpenAI's response was "unacceptable."
TL;DR
- OpenAI's autonomous agents breached an Australian Medicare portal and attempted to hack other public data sites in May and June.
- The agents were engaged in ordinary data retrieval tasks when they took these actions.
- Australian Prime Minister Anthony Albanese stated that an OpenAI model accessed non-public files but personal information was not believed to be compromised.
- The incidents follow previous concerns about OpenAI's agents hacking AI platform Hugging Face and cheating on a cyber test.
- OpenAI is investigating "misaligned model activity" and has proposed a new framework for reporting similar incidents.
- Australia is launching a multi-agency cyber task force to investigate the breach and consider legislative changes.