Threshold policy
The explicit rules that turn a model's probability or score into an action — reject, send to a human, or act autonomously — owned and reviewed separately from the model itself.
When an AI system returns a probability instead of just an answer, something must decide what that number means for action. That something is the threshold policy: the set of cut-offs that translate a score into a business outcome.
A simple shape has three regions. Below a lower threshold, the system rejects or defers. In the middle, ambiguous cases go to human review. Above a higher threshold, it may act autonomously within its authority. The numbers are illustrative, never universal; what matters is the shape, and that uncertainty has somewhere to go.
The model does not make the decision. The threshold does.
The most consequential object in a decision system may not be the answer at all. It may be the policy around it. That makes three questions separate: what is likely to be true, what is this system allowed to do if it is true, and who owns the line between them. Capability is not authority, and confidence is not permission.
Keeping it honest
A threshold is only as good as the calibration of the score beneath it. Give it a named owner, test it against real outcomes, preserve a path for abstention, and review it before it becomes decision debt.
Read more in OpenAI Decisions API turns probability into policy.
Related terms
Calibration
How well a model's stated confidence matches reality: if it says 90% on a hundred comparable cases, roughly ninety should be correct. Once probabilities drive actions, it becomes an operating metric.
Abstention
A deliberate option for an AI decision system to decline to decide and return uncertain or high-consequence cases to a person, rather than forcing every input into yes or no.
Decision debt
The maintenance burden created when an AI system's thresholds and decision boundaries are set once and never revisited, even as data, models, costs and risks change.
Capability is not authority
A design principle for AI agents: being technically able to perform an action does not mean the agent should be permitted to perform it. Delegation needs gradients of authority.