Open letters shape AI policy conversations—but most fizzle after the press cycle. Here’s a short, practical template to turn signatures into outcomes.
For context on current debates, see Simon Willison’s roundup of recent AI open letters: Open letters. Use the guide below to make your next one specific, verifiable, and hard to ignore.
Why open letters still matter
They create public coordination across fragmented stakeholders and set a visible benchmark for progress. Done well, they compress complex risks into clear, testable asks.
Ground your asks in recognized frameworks. Two useful anchors: the NIST AI Risk Management Framework and the OECD AI Policy Observatory.
The 7-step checklist
- Audience: Name the decision-maker with authority (e.g., agency, standards body, or company board).
- Specific ask: One to three actions, each measurable and time-bound.
- Evidence: Cite data, incidents, or benchmarks that justify urgency and feasibility.
- Scope of impact: Who benefits, who bears cost, and how risks are mitigated.
- Accountability: Define success metrics, reporting cadence, and an independent verifier.
- Signatories: Mix expertise—policy, technical, civil society, and affected users—not just celebrities.
- Follow-through: State the next step if ignored (e.g., hearings request, shareholder proposal, or standardization ballot).
Copyable structure (use as a template)
- Title: A clear outcome (e.g., “Adopt baseline red-teaming for foundation models by Q1”).
- Problem in one paragraph: Concrete harm or risk, with a stat or case.
- Three asks: Each with owner, deadline, and metric.
- Evidence: 3–5 citations to standards, incident databases, and research.
- Implementation notes: Cost, staffing, and sequencing.
- Verification plan: Who measures, when, and how findings are published.
- Signatories: Ordered by role and expertise; include affiliations and conflict disclosures.
Measure impact, not virality
- 30 days: Meetings secured with target decision-makers; public acknowledgement.
- 90 days: Policy drafts, RFPs, or internal memos referencing your asks.
- 180 days: Audits launched, budgets allocated, or standards ballots initiated.
Common pitfalls to avoid
- Vague asks (“do better on safety”) without owners, budgets, or deadlines.
- Signatory monoculture that weakens legitimacy and press interest.
- No escalation path if ignored—silence becomes the default response.
Examples worth studying
- The Future of Life Institute’s 2023 letter on pausing giant AI experiments—clear, time-bound framing: read it.
- The Center for AI Safety’s one-sentence risk statement—concise problem definition: view it.
Takeaway
Make your AI open letter impossible to sidestep: pick one decision-maker, three measurable asks, and a public verification plan. Everything else is optional.
Enjoy pieces like this? Get one actionable AI nugget in your inbox, weekly. Subscribe to The AI Nuggets.
