Summary
OpenAI warns that its future AI models will likely pose a high cybersecurity risk due to their increasing capabilities. These models could help more people carry out cyberattacks, and OpenAI is taking steps to prepare for and mitigate these risks.
Key Facts
- OpenAI says its upcoming AI models could pose a "high" cybersecurity risk.
- Recent models have shown significant improvements in cybersecurity capabilities.
- GPT-5 scored 27% and GPT-5.1-Codex-Max scored 76% in a security test.
- "High" risk is the second-highest risk level, below "critical."
- OpenAI is preparing for new models that could reach "high" cybersecurity levels.
- OpenAI is working with other labs on cybersecurity threats through the Frontier Model Forum.
- The company plans to establish a Frontier Risk Council for cybersecurity collaboration.
- OpenAI is testing a tool called Aardvark to identify security gaps in products.
