Reinforcement learning trains a policy through reward and interaction with an environment, not through a fixed labelled dataset.
What the base framework establishes
The approach is defined by feedback rather than by examples: an agent acts, the environment responds, and a reward shapes the behaviour over time. The framework names it as a distinct approach because the evidence is temporal and behavioural. A single accuracy figure cannot describe it; the record needs the reward design, the environment, the exploration policy and the behaviour observed after deployment. The clause text is not in the public sample. The page states the approach's place in the framework and does not attribute wording to the clause.
Why it matters to an AI management system
Reward design is where reinforcement systems fail in a way that looks like success. An agent optimises what the reward measures, including the parts nobody intended. Governance needs the reward definition and a record of behaviour outside the training environment, or it is signing a blank cheque.
The public draft summary names no change to 7.6 Reinforcement machine learning. The final text could still touch it.
Evidence a reviewer samples
| Category | Artefact |
|---|---|
| Reward specification | The reward terms, weights and the behaviour they encourage. |
| Environment record | The training environment and how it differs from production. |
| Behaviour review | Observed behaviour after deployment, including unintended strategies. |
One question to prepare for. What does the reward actually reward, and what does it ignore?
Common gaps
- The reward is a single scalar with no record of what it excludes.
- The production environment differs from training and the difference is not assessed.
- No one reviews the agent's behaviour after go-live.
Prepare before publication
Document the reward and the environment gap, and add a post-deployment behaviour review with a named reviewer.
Dates that matter
No separate date attaches to 7.6; the record dates are the stage date, the planning estimate and the CEN expectation stated above.
Cross-references
6.2 Task · 7.1 General · 8.7 Operation
Source
Source for 7.6: ISO/IEC 23053:2022, clause 7.6. The public sample stops before 7.6, so this page relies on the public contents and the public summary and does not quote the clause. The two official records are linked on this page so any line can be checked.
Not legal advice; does not certify. A reading of the framework and the evidence a reviewer commonly samples, not a substitute for the standard or your auditor's rules.