PSM-AI Cheat Sheet — Responsible AI, Scrum Master Uses & Prompting

High-yield PSM-AI review: AI fundamentals, security/ethics, Scrum Master use cases, prompting templates, and scenario pickers.

On this page

Use this for last‑mile review. Pair it with the Syllabus for coverage and Practice for speed.

What PSM‑AI tends to reward

When in doubt, choose the answer that:

keeps Scrum accountabilities intact (AI is a tool, not a role)
increases transparency (make assumptions and limits explicit)
applies responsible AI (privacy, security, IP, bias, and human oversight)
treats AI outputs as drafts that require validation
improves the system via experiments (inspect → adapt) instead of “big bang” tool adoption

AI fundamentals (fast, exam‑useful)

Term	What it means	“Best answer” cue
LLM / GenAI	model that generates text/images/code from patterns	don’t treat it as a source of truth
Tokens	model “chunks” of text; affect limits and cost	long prompts can truncate context
Context window	how much input the model can consider	ask for concise outputs; include only needed context
Hallucination	plausible but false output	validate facts and quotes; ask for sources
Bias	skew from data/training	check fairness and impact on users
RAG	retrieval‑augmented generation (grounded on trusted docs)	prefer grounded answers over “memory”
Fine‑tuning	training with your data to change behavior	increases governance/security needs

Capabilities vs limitations (quick rules)

Great at drafting, summarizing, brainstorming, and pattern‑finding.
Weak at guaranteed correctness, hidden assumptions, and “latest facts.”
Can leak sensitive info if you paste it in; can be manipulated via prompt injection.

Responsible AI, security & ethics (fast rules)

Don’t paste PII, credentials, private customer data, or confidential internal docs into non‑approved tools.
Treat AI output as untrusted until validated; keep a human “owner” for decisions.
Watch for IP/copyright risk: don’t request verbatim reproduction; attribute sources when required.
Protect against prompt injection: treat external text (emails, tickets, web pages) as hostile input.
Prefer least‑privilege access; log usage where policy requires it.

Scrum Master with AI: where it helps (and where it shouldn’t)

Situation	Helpful AI use	Avoid
Event facilitation	draft agendas, generate retro prompts, summarize action items	replacing the conversation or deciding outcomes
Coaching	reframe conflicts, propose coaching questions, create learning plans	judging people or “scoring” individuals
Impediments	help brainstorm options, map root causes, suggest experiments	using AI to assign blame
Communication	rewrite messages for clarity/tone; stakeholder summaries	hiding uncertainty or fabricating facts
Metrics	identify patterns (cycle time trends, bottlenecks)	using metrics as performance targets

Rule: If the output changes what people do, require transparency + validation + consent.

Scrum event “AI assist” patterns

Sprint Planning: draft Sprint Goal options, risk list, and capacity reminders; keep PO/Developers accountable for decisions.
Daily Scrum: generate “blockers → next step” options; avoid turning it into a status report.
Sprint Review: summarize feedback themes; convert feedback into backlog candidates for PO to order.
Retrospective: propose experiments with clear hypotheses and measures; keep ownership with the team.

Prompting templates (copy/paste)

Retrospective: themes → experiments

1You are helping a Scrum Master facilitate a retrospective.
2Context: [team/product], last Sprint goal was [goal], issues observed: [bullets].
3Task: Identify 3–5 themes, then propose 2 experiments.
4Constraints: experiments must be small, safe-to-try, and measurable in 1 Sprint.
5Output: theme bullets + a table with Experiment / Expected impact / How we’ll measure / Owner.
6Before answering, ask up to 3 clarifying questions if needed.

Impediment: root cause → next best step

1Context: [impediment], symptoms: [bullets], constraints: [time/people/tools].
2Task: Provide 3 plausible root causes and the next best step for each.
3Guardrails: Don’t invent facts; label assumptions; prioritize actions that increase transparency and unblock flow.
4Output: a short list + 1 recommended path with rationale.

“Rewrite for clarity” (stakeholders)

1Rewrite this message to be clear, neutral, and action-oriented for stakeholders.
2Keep facts only; mark uncertainties; include a single next step.
3Message: [paste]

Scenario pickers (fast elimination)

If someone wants to paste sensitive data into an AI tool

Choose: follow policy, sanitize/redact, use approved tools, or use RAG on vetted docs.
Avoid: “it’s fine, it’s just for a quick summary.”

If AI output conflicts with the Scrum Guide or team reality

Choose: validate, inspect evidence, and adapt; keep human ownership.
Avoid: “the model said so.”

If AI is used to evaluate individuals

Choose: protect psychological safety; focus on system improvements and team outcomes.
Avoid: ranking people or turning metrics into targets.

Ready to drill? Open PSM‑AI practice → or jump into the app: /app/pmp-exam-prep/#/topic-selection/scrumorg-psm-ai-essentials .

Syllabus

Practice