OpenAI Warns Upcoming Models Approach High Cybersecurity Risk Under Preparedness Framework

Summary

OpenAI published updates to its Preparedness Framework and GPT-5.2 system card in the week of December 9-12, 2025, revealing that its latest models are approaching — but have not yet crossed — the threshold for "high" cybersecurity risk. The disclosure is notable for its transparency about dual-use capabilities: GPT-5.2 and GPT-5.2-Codex demonstrate significantly stronger cybersecurity capabilities than any previous OpenAI model, with performance showing sharp jumps in capability across successi