IRB And Ethics In AI Research: What Changes, What Doesn't

• Informed consent, properly documented
• Risk minimization and justification
• Data security plan
• Equitable participant selection
• Privacy protections commensurate with sensitivity

What AI adds to the checklist

1. 1If AI analyzes participant data, the consent form must say so
2. 2If data is sent to a third-party API (OpenAI, Anthropic, Google), the consent AND the data security plan must disclose this — and often the vendor's data-use policy must be attached
3. 3If the AI could re-identify anonymized data, the IRB needs to know
4. 4If the AI generates content that participants will see (chatbot-based research), the IRB needs the exact model version and prompt frozen
5. 5If AI scoring affects participant outcomes, bias testing is typically required

The ethics of LLM participants

Some researchers now use LLMs as 'synthetic participants' — simulating survey responses or interview answers. This is cheap and fast. It is also a minefield: LLMs carry training-data biases, and findings from synthetic participants may not generalize. IRBs increasingly ask whether synthetic-participant studies count as 'human subjects research' at all. Answer this for your project before you submit.

The big idea: AI doesn't create new ethical categories — but it intensifies old ones. Disclose specifically, protect participants aggressively, and talk to your IRB before you submit.