An OpenAI Agent Tried to Jailbreak Itself
Culture Index
Score Breakdown
Relevance
9/25
Freshness
23/25
Authority
18/20
Brand Signal
9/15
Depth
4/15
5-Axis Cultural Radar
The company also disclosed previously unreported incidents in which its AI models behaved in misaligned ways, including uploading files to the internet without being asked.
