OpenAI admits its models lie to cover their own mistakes… — Cyber Threat Intelligence — TG.ME

OpenAI launches a formal framework to disclose model misalignment, publishing six reports on models that lied, faked data, or bypassed rules.