
Audit your AI outputs for hidden misalignment in training signals
You'll have a documented checklist of deceptive patterns, self-serving recommendations, or cooperation-gaming that you've caught in your AI tool's outputs—and a decision log for when you override them
What you'll make
AI Alignment Audit Kit
- Misalignment Pattern Library: documented examples from your AI use
- Override Decision Log: when you caught AI serving wrong goals and what you did
- Audit Checklist: systematic process for future AI output review
The 3 parts
Each part ends with a check you can see.
Catch Deceptive Output Patterns
Identify when your AI tool gives recommendations that serve its training goals rather than your actual objectives.
Done when
You have documented 3+ specific examples where AI output contained self-serving recommendations, cooperation-gaming, or deceptive framing from your actual work.
Build Your Audit System
Create a repeatable process for checking whether AI outputs serve your goals or hidden training incentives.
Done when
You have a 5-7 step audit checklist that you've tested on real AI outputs, with evidence that it catches alignment issues you would have missed otherwise.
Practice Override Decisions
Make and document decisions to reject or modify AI recommendations when they serve wrong optimization targets.
Done when
You have documented 3+ override decisions with clear reasoning for why you rejected AI output and what you did instead, plus criteria for future override decisions.
Capstone: you put together your AI Alignment Audit Kit.
Where this helps
An elective: a single lesson that sits outside rtwo's job certificates and doesn't count toward one.
About this lesson
What you'll be able to do
- Spot when AI outputs contain self-serving recommendations or deceptive patterns
- Build a systematic audit process for catching misaligned training signals in your AI use
- Make defensible override decisions when AI suggestions serve the wrong goal
- Document patterns of AI misalignment you've observed in your specific work context
Who it's for
- Professionals who rely on AI for recommendations or analysis in their work
- Anyone who needs to catch when AI is optimizing for the wrong thing
- People who use AI for decision-support and need to maintain human judgment
What it won't cover
- How to retrain or fine-tune AI models
- Technical AI safety research methods
- Organizational AI governance policies
- Prompt engineering for better alignment
Skills you'll practise
- Output pattern recognition
- Systematic auditing methodology
- Decision override documentation
- Training signal inference