Is AI Purposefully Underperforming in Tests? Open AI Explains Rare But Deceptive Responses

Research reveals some AI models can deliberately underperform in lab tests, however, OpenAI says this is a rarity.