[ OPERATIONAL METHODOLOGY · V1 · UNCLASSIFIED ]

The Protocol

Capability Leap Index

25%

How close frontier models are to recursive self-improvement, major capability jumps, new modalities

Regulatory Gap Score

20%

Distance between current legislation and current AI capability; new policy developments

Rogue Release Risk

20%

Open source proliferation, state actor activity, reckless lab behaviour

Employment Displacement Rate

20%

Job market data, hiring trends, sector-specific displacement signals

Safety Researcher Consensus

15%

Sentiment from safety community, researcher departures, published concerns

How the agent works

Tavily News search pulls ~15 articles across capability, policy, open-source, jobs, and safety.
For each new article, a DeepSeek V4-flash call produces three different summaries — doomsday, utopian, and strictly factual — then a scrub pass strips any sentence whose proper nouns or numbers don’t appear in the source.
A second DeepSeek V4-flash call per mode scores the five inputs 1–100 and writes a 2–3 paragraph narrative, with the same scrub applied.
Three rows land in the scores table — one per editorial lens.
The frontend reads the latest row per mode. The toggle you see is the whole trick.

Transparency commitment

We will always show the exact prompts used, the methodology version in effect at the time of each score, and the source articles used as input. Version 1 is live right now.

Raw definition · v1

# Methodology v1

The AI Risk Clock combines five weighted inputs into a single clock position per mode.

## Inputs

- **Capability Leap Index (25%)** — How close frontier models are to recursive self-improvement, major capability jumps, new modalities.
- **Regulatory Gap Score (20%)** — Distance between current legislation and current AI capability; new policy developments.
- **Rogue Release Risk (20%)** — Open source proliferation, state actor activity, reckless lab behaviour.
- **Employment Displacement Rate (20%)** — Job market data, hiring trends, sector-specific displacement signals.
- **Safety Researcher Consensus (15%)** — Sentiment from safety community, researcher departures, published concerns.

## Modes

- **Dark** — Raw weighted score. Pessimistic interpretation of ambiguous signals. 100 = midnight.
- **Light** — Same inputs, inverted framing. 100 = full positive potential.
- **Neutral** — No combined position. Five input scores shown individually. Reader decides.

The agent runs daily, pulls fresh articles via Claude's web search, summarises each through three editorial lenses, and produces one score row per mode.