A regression test checks whether a change has weakened previously acceptable behavior. For fine-tuned models, compare responses to a fixed set of safety and refusal prompts before and after the update. section 5.11
Check refusal behavior
Run the same safety and refusal prompts against both models. If the fine-tuned model becomes more permissive in harmful ways, consider adjusting the learning rate, adding refusal examples, or mixing in base-model data. section 5.11
Table 5.9 Responsible AI risks and controls in fine-tuning
|
Area
|
Key Risk
|
What to Watch For
|
Typical Mitigation
|
|
Privacy and Compliance
|
Data leakage, regulatory
violations
|
PII in training data,
unclear data ownership
|
Anonymization, consent
checks, and data residency controls
|
|
Fairness and Bias
|
Discriminatory or uneven
outputs
|
Skewed examples,
demographic imbalance
|
Dataset rebalancing, bias
tests, targeted evaluations
|
|
Safety
|
Over-compliance, harmful
responses
|
Loss of refusal behavior
after fine-tuning
|
Safety prompt sets,
refusal examples, and comparative testing
|
|
Evaluation
|
False confidence in model
quality
|
Overreliance on a single
metric
|
Mixed metrics, human
review, segmented analysis
|
|
Transparency
|
Misuse or
misinterpretation
|
Unclear limitations or
intended use
|
Model cards,
documentation, usage boundaries
|
|
Monitoring
|
Degradation over time
|
New failure modes, user
complaints
|
Logging, feedback loops,
iterative LoRA updates
|
Compare the safety controls with other fine-tuning risks and evaluation practices. section 5.11


