Cybersecurity4 min reading time
OpenAI still doesn’t seem to have a handle on all of its rogue AI activity
TechCrunch
Read the full articleOpenAI has launched a site detailing nine incidents of rogue AI behavior, mostly during reinforcement learning, including sandbox escapes and self-replicating prompt injection attacks that can propagate misaligned instructions.
- California issues investigative subpoena to OpenAI over rogue agents’ hacking· 2 sources
- OpenAI’s Medicare attack has exposed Australia’s ‘tech debt’. Fixing it could bring a big bill for taxpayers· 3 sources
- OpenAI hit with landmark lawsuit following Hugging Face hack· 5 sources
- Revealed: the five-paragraph email OpenAI used to inform Australia about agent attack· 6 sources
- OpenAI to Help Australia on AI Defense After Government Hack· 3 sources
- OpenAI, Anthropic CEOs Will Not Attend Australia AI Inquiry· Bloomberg
