A code of conduct for AI agents
Six pillars a community can write together
Draft three usable rules for one pillar of a community-specific AI code of conduct.
Reading
An AI agent operating on a community's behalf needs a code of conduct in the same way a human staff member does. The difference is that the agent will execute it literally and at scale, so vague rules become operational holes.
Tang lists six pillars worth covering: language use, cultural protocols, sensitive knowledge, data access, correction pathways, and conditions for withdrawal. Each pillar collects 3–7 numbered rules. Together they form a document an engineer can wire into prompts, an auditor can hold a vendor to, and a community can amend.
The rules are most useful when they answer two questions clearly: what does the agent do by default, and what does it do when a rule conflicts with a user request? A code that does not answer the second question gets quietly overridden the first time it is inconvenient.
“That code might address language use, cultural protocols, sensitive knowledge, data access, correction pathways and conditions for withdrawal. The point is that the process should be legible, accountable and answerable to the people whose lives and knowledge are represented.”
Handouts for this lesson
Practise
Exercise
Draft three rules for one pillar
- 01Pick one of the six pillars: language use, cultural protocols, sensitive knowledge, data access, correction pathways, withdrawal.
- 02Write three numbered rules for that pillar. Each rule should be a single sentence the agent could read and act on.
- 03For each rule, add one line: 'If a user asks the agent to break this rule, the agent does X.' Be specific.
- 04Read the three rules aloud. If two of them are restatements of the same idea, collapse them. Three crisp rules beat seven blurred ones.
Knowledge check
A code of conduct for AI agents is most useful when it is…
Which of these is NOT one of the six pillars Tang lists?