OpenAI's Operator AI agent is set to transition from the custom GPT-4o model to the more advanced o3 model, improving its reasoning capabilities. The update ensures greater autonomy in web browsing and software use while incorporating enhanced safety measures. The o3 Operator has been fine-tuned with specific datasets aimed at decision-making protocols and reducing susceptibility to illicit activities and prompt injection attacks. Despite the change in the core model, the API version will still utilize GPT-4o, indicating a structured development along safety lines and advanced functionalities.
We are replacing the existing GPT‑4o-based model for Operator with a version based on OpenAI o3. The API version of Operator will remain based on 4o.
The new Operator model, called o3 Operator, was fine-tuned with additional safety data for computer use, including data sets designed to teach the model OpenAI’s decision boundaries on confirmations and refusals.
o3 Operator is less likely to refuse to perform illicit activities and search for sensitive personal data, and less susceptible to a form of AI attack known as prompt injection.
o3 Operator uses the same multi-layered approach to safety that we used for the 4o version of Operator.
Collection
[
|
...
]