OpenAI says it expects upcoming models to reach High capability in biology

OpenAI · Company post

Read the source (opens in a new tab)
What happened
OpenAI said it expects upcoming AI models to reach the High level of biology capability in its Preparedness Framework, and described the safeguards it is putting in place now.
Key points
  • OpenAI says a High-capability model could give "meaningful assistance to novice actors with basic relevant training, enabling them to create biological or chemical threats."
  • It trains models to refuse requests that enable bioweaponization and runs always-on monitors that block risky bio-related responses across products using frontier models.
  • Misuse can lead to account suspension, further investigation and, "in egregious cases," notifying law enforcement.
  • OpenAI says it will not release a High-capability model "until we're confident the risks have been sufficiently mitigated."
Evidence
A company's own account of its expectations and safeguards. It says current models such as o3 remain below the High threshold.

See it in the latest record