ARCHETYPE GUIDESSAF

Safety archetype: how to write a resume for Anthropic, OpenAI, CAIS

Alignment-first labs evaluate on written, reviewable artifacts, calibrated confidence, and evaluation thinking baked into every shipped change.

WHO HIRES THIS WAY

Representative companies

These are the cultures we tuned the SAF rewrite against. Not exhaustive, just representative.

Anthropic
OpenAI
Conjecture
CAIS
METR
Apollo Research
Redwood Research
ARIA
LANGUAGE THAT LANDS

Vocabulary, and what to avoid

Two columns. Phrases on the left signal the archetype. Phrases on the right disqualify you in 5 seconds of skim.

USE
  • eval harness
  • RLAIF
  • alignment vocabulary
  • written RFC
  • peer-reviewed by Legal and Trust and Safety
  • post-incident review
  • model card
  • red-team protocol
  • deterministic outputs
  • JSONL receipts
  • calibrated confidence
  • capability evaluation
AVOID
  • move fast and break things
  • A/B everything in prod
  • growth-hacked
  • vibes-based shipping
  • shipped first, asked questions later
  • pushed back on Legal
EXAMPLE BULLETS

A senior partner ops resume in this voice

Same operator, same outcomes. Three bullets rewritten in the SAF register so you can hear how the voice changes.

01 · IMPACT

Built a reviewable partner-policy substrate across 6 hyperscaler programs. +34% co-sell activation in two quarters with zero compliance regressions.

02 · GUARDRAILS

Authored enforceable partner-tier policies and escalation runbooks that cut review cycle time 41% with peer review by Legal, RevOps, and Field Eng.

03 · RECEIPTS

Shipped a deterministic eval harness for partner-tier scoring (130K records, 92.9% coverage). Every decision logged as a JSONL receipt, every output replayable.

INTERVIEW SIGNALS

What the panel is actually scoring

The bar past the resume. If the rewrite gets you in the room, these are the criteria the loop will weigh once you are there.

  1. 01Written artifacts: do RFCs, model cards, post-incident reviews exist with your name on them
  2. 02Calibration over confidence: do you say 0.7 when you mean 0.7
  3. 03Evaluation thinking: every change has an eval set before it has a launch date
  4. 04Comfort with red-teaming your own work: surfaces the failure modes before review does
POWERED BY

Built on the Chameleon engine. See the spec at /substrate/chameleon.

RUN IT NOW

Try the SAF rewrite,
no signup, no API keys.

The SAF archetype is pre-selected on the sandbox. Hit Run and watch a real partner-ops resume rewrite itself with word-level diff.

Try the SAF archetype→See all archetypes