FLAME Framework

THE AI DECISION AUDIT.

FLAME is the framework I built to help leaders evaluate AI-driven decisions before they become operational, ethical, or reputational liabilities. Five dimensions. One clear standard for accountability in the age of intelligent systems.

FUTURE

Does this decision account for long-term consequences, not just immediate outcomes?

LEADERSHIP

Is there a human leader who owns this decision and can be held accountable?

ANALYTICS

Is the data behind this decision accurate, unbiased, and fit for purpose?

MORALITY

Does this decision align with the moral values of the institution and the communities it serves?

ETHICS

Does this decision meet the ethical standards expected of responsible AI deployment?

How It Works

THREE STEPS. TOTAL CLARITY.

IDENTIFY THE DECISION

Define the AI-driven decision under review. Who made it? What data drove it? What outcome did it produce?

AUDIT THE FIVE DIMENSIONS

Run the decision through each FLAME dimension. Score it. Surface the gaps in accountability, data quality, and ethical alignment.

ACT ON THE FINDINGS

Use the audit results to course-correct, strengthen governance, assign accountability, and build institutional capacity to govern AI responsibly.

Who It's For

BUILT FOR LEADERS WHO ARE ACCOUNTABLE.

Executives & C-Suite

CEOs, COOs, and board members who need a clear standard for evaluating AI decisions before they escalate into institutional risk.

Compliance & Risk Teams

Risk officers and compliance leaders building governance frameworks that withstand regulatory scrutiny.

Technology Leaders

CTOs and technology directors who deploy AI systems and need a structured way to ensure those systems meet organizational and ethical standards.

Public & Institutional Leaders

Government officials, university administrators, and nonprofit leaders whose AI decisions carry public accountability.

READY TO RUN YOUR FIRST AUDIT?

The FLAME AI Decision Audit is available now. Start evaluating your AI-driven decisions with a framework built for leadership accountability.

Launch the Audit Read the Book

70/100

May 1, 2026 — FLAME Conditional

First audit to reach FLAME Conditional. Applied all governance improvements: Extended Unavailability Protocol published, Christian Calix tertiary Vercel access documented with recusal language committed, six governance documents published at GitHub, consistency test completed and published with 1-point variance, bilingual English and Spanish support deployed, returning user flow with PDF upload added, formal secondary appeals process with binding 14-day review, named legal counsel. Leadership scored 16/20. Analytics scored 14/20. Ethics scored 13/20 for the first time without a scoring error.

66/100

May 1, 2026 — FLAME In Development (Consistency Test B)

Consistency Test B. Same decision, same governance infrastructure as Test A. Total variance: 1 point (67 vs 66). Leadership dropped 3 points on harder follow-up questions about verification sequences and cooling-off periods. Analytics gained 3 points with bilingual accessibility and consistency threshold framing. Future, Morality, and Ethics held within 1 point. Consistency confirmed within acceptable threshold.

67/100

April 30, 2026 — FLAME In Development (Consistency Test A)

Test A of consistency test. Applied governance infrastructure improvements: Triage Protocol v2, Wilmer Toro with documented emergency authority and full credential access, Christian Calix as independent tiebreaker, Doug Kim of Kim Lahey and Killough named as legal counsel, three Google Alerts active, methodology review outreach sent to AI Now Institute and Algorithmic Justice League, bilingual English and Spanish support deployed, PDF report generation and returning user flow added. Leadership scored 16/20.

66/100

April 29, 2026 — FLAME In Development

Applied recommendations from the 59/100 audit: added response summarization, minimum 2 exchanges per dimension, probing limits, and report generation retry logic.

59/100

April 28, 2026 — FLAME In Development

Applied recommendations from the 48/100 audit: published Terms of Service and Privacy Policy, implemented anonymized session logging, added active harm detection, and documented escalation procedures at flame@rodrik.io.

48/100

March 25, 2026 — Not FLAME Certified (Launch)

Launch score. No Terms of Service, no Privacy Policy, no session logging, no formal escalation path, basic keyword-only bias detection. Published as-is.

Priority gaps remaining: Priority gaps remaining: Calix formal written agreement pending, fourth operational emergency contact not yet identified, independent adversarial testing of harm classifier not yet conducted, external methodology review responses pending from AI Now Institute and Algorithmic Justice League.

Current status: FLAME Conditional · Consistency test completed May 1, 2026. Variance: 1 point (acceptable). Two post-audit improvements identified. Score: 70/100 · May 1, 2026 · Target: FLAME Certified (85+) by Q4 2026