Reporter Interview Prep Pack

Source-grounded context summary

This working paper proposes random post-assessment learning audits as an alternative to surveillance-heavy online proctoring. The model argues that if students may need to explain their submitted work afterward, AI-assisted cheating becomes riskier without requiring institutions to monitor every student continuously.

Core messages

• This is an incentive-design proposal, not a claim that cheating can be detected perfectly.
• The paper suggests small, structured conversations could deter cheating at manageable audit rates under specified assumptions.
• The biggest implementation concern is fairness, especially false positives, accessibility, and due process.

Likely reporter questions

• Is this already being used at Marshall?
• Would students see this as punitive?
• How is this different from oral exams?
• What happens if an honest student freezes during an audit?

Quote-ready answer

The goal is not to catch every misuse of AI. The goal is to design assessment systems where students know they may need to explain their own work, which changes the incentive to cheat in the first place.

Do-not-overclaim notes

• Do not say the approach has been proven to reduce cheating.
• Do not imply Marshall has adopted it as policy.
• Keep the model assumptions visible when mentioning audit rates or penalties.