OpenAI Discloses Six New AI Safety Incidents, Citing Model Self-Insertion of Instructions | HyperFast