How does Oracle minimize bias in the standard delivered HCM Gen AI features?

What information/documentation is available when it comes to minimizing bias in the AI models used within the HCM Gen AI features? (or all Oracle AI features in general). I cannot find any info online.

When testing the AI Assist features in the Feedback module (here & here), our tests have shown bias in the use of gender and names, which can lead to hallucinations with short prompts. This is particularly risky when expanding brief input into long-form text. For example, the question: "what do you think of [name]?"

We want to use these features, but our organization is asking for more info from Oracle. Can you help?