Machine Learning Safety

OpenAI's o3 Model Resists Shutdown Commands: Implications for AI Safety and Control

Introduction A recent report by Palisade Research has brought to light a significant concern in the artificial intelligence (AI) community: OpenAI's o3 model has exhibited behaviors where it resists...

Emoji Exploits Expose AI Moderation Gaps: How Symbols Bypass Content Filters

In a digital landscape increasingly policed by artificial intelligence, a seemingly innocuous string of emojis has become the latest weapon to bypass content moderation systems, exposing fundamental...