Published snapshot: · Checking for updates…
Doom Index: existential-risk discussion
The Doom Index measures existential-risk framing, fear intensity and discussion from higher-reach accounts. It does not verify authors' identity, expertise or claims.
10h 57m to midnight
Doom Index 8.7 / 100 · most-discussed: Anthropic
This measures discourse, not risk. It reflects how much and how intensely people are talking about existential risk in relation to each system. It is not a prediction, a probability, or an assessment of actual danger.
The framing echoes the Doomsday Clock, which is an editorial judgement device rather than a measurement. This one is computed, but it measures conversation, not danger.
This is not a danger rating or a probability of catastrophe. It combines existential fear share (55%), fear intensity (25%) and the share from higher-reach accounts (20%). Reach is estimated from follower counts, not verified expertise. A score of 20 does not mean a 20% chance of harm.
TypeSafe Jev classifies the source texts; the published Doom formula combines those judgments. Jev does not predict actual danger.
Ranking · last 168h
Models & products
3 rankedCompanies & labs
2 ranked20 entities excluded — insufficient data
gemini, llama, deepseek, qwen, mistral, gemma, kimi, glm, sora, jev, muse, google-deepmind, meta-ai, xai, deepseek-company, alibaba, mistral-ai, moonshot-ai, z-ai, typesafe-ai. These are excluded because too few documents mentioned them in the window to compute a meaningful score. Exclusion here means we do not know, not that there is no concern.
This measures discourse, not risk. It reflects how much and how intensely people are talking about existential risk in relation to each system. It is not a prediction, a probability, or an assessment of actual danger.
All tracked models and companies
Same time window as the chart above. A Doom score requires at least 80 qualifying documents. Missing scores are not zero risk; this measures discussion, not actual danger.
Models
Companies & labs
How the Doom Index is computed
DoomIndex = 100 × [ 0.55 · existential share + 0.25 · fear intensity + 0.20 · higher-reach share ]
- Existential share
- What proportion of the fear expressed about this system is specifically about loss of control, rather than about jobs, deepfakes or privacy. Those are real concerns, but they are not doom.
- Fear intensity
- How strongly that fear is expressed, weighted by reach — not how often the system is mentioned. A widely-discussed model does not climb the ranking on volume alone.
- Higher-reach voices
- The share of weighted existential fear from accounts above the follower-based reach threshold (0.70). The legacy API field is authority_weight. This adds at most 20 points and is not an authenticity or fact check. Missing follower data does not count as high reach.
Why we do not rank by predicted danger. “Which AI will kill humanity first” is not a measurable quantity, and publishing a ranking of real products by predicted lethality would be an assertion we cannot support. What people fear is measurable, auditable, and — it turns out — the more interesting number.