OpenAI Reports Cybersecurity Capabilities Advance While Biological Threats Remain High
OpenAI's latest GPT-5.2 system card reveals the model achieved 82% success on professional cybersecurity challenges and continues to exceed expert baselines on biological threat evaluations.
Read next
Anthropic’s New AI Framework Targets Downstream, Man Made Perils
Anthropic says AI risk theory is tied to man-made perils.
US Says Natural Hazard Models Need AI, Despite Challenges
Leveraging private market expertise is one solution the US government is mulling.
