At a Glance · AI-generated

OpenAI has undertaken sustained efforts to evaluate and mitigate risks in large language models through multiple safety initiatives spanning September 2023 to August 2025. The company launched a Red Teaming Network in September 2023 and subsequently advanced red teaming methodologies combining human and AI testing by November. OpenAI developed specialized tools including an early warning system for biological threats and detection methods for misbehavior in reasoning models, while publishing its comprehensive approach to frontier risk mitigation. In August 2025, OpenAI and Anthropic jointly released findings from coordinated safety evaluations, demonstrating collaborative progress on industry-wide safety practices.