Source index

Source index

Everything this manual cites, newest first. Each entry records how direct the source is, so a reader can weigh a peer-reviewed report differently from a newspaper writing up an interview.

Revised
How to read this

A summarized position is our sentence describing what a source argues. A direct quotation is text checked word-for-word against the original. Nothing in this manual is presented as a quotation unless it has been checked, which is why most entries below are summaries.

Register

14 sources — 2 interview, 5 primary, 2 secondary, 2 report, 3 statement.

  1. 01

    Summary Yoshua Bengio says the field is building systems it does not yet know how to control.

    Yoshua Bengio — Founder, LawZero; Professor, Université de Montréal

    Bloomberg Businessweek Interview

    Cited on The loss-of-control mechanism, step by step

  2. 02

    Summary Dario Amodei states that we have no precise understanding of why models choose the words they do.

    Dario Amodei — CEO, Anthropic

    The Urgency of Interpretability Primary

    Cited on Interpretability: reading what a model is doing internally

  3. 03

    Summary Demis Hassabis puts human-level AI five to ten years out, with meaningful evidence already in play.

    Demis Hassabis — CEO, Google DeepMind

    CNBC Interview

    Cited on AGI timeline forecasts: what the labs and surveys say

  4. 04

    Summary Yoshua Bengio warns that a loss of control could follow from AI systems developing their own drive to survive.

    Yoshua Bengio — Turing Award laureate; lead author, International AI Safety Report

    AFP, via TechXplore Secondary

    Cited on Race dynamics and their effect on safetyThe loss-of-control mechanism, step by step

  5. 05

    Summary World Economic Forum projects roughly 92 million roles displaced against 170 million created globally by 2030.

    World Economic Forum — Future of Jobs Report 2025

    World Economic Forum Report

    Cited on Labor displacement: what the data actually shows

  6. 06

    Summary Communications Fraud Control Association reports that as little as three seconds of audio can clone a person’s voice.

    Communications Fraud Control Association — Five Ways to Protect Your Voice from AI Voice Cloning Scams

    CFCA Report

    Cited on Deepfake and voice-clone scams: household defenses

  7. 07

    Summary Geoffrey Hinton estimates a 10–20% chance that AI drives human extinction within thirty years.

    Geoffrey Hinton — Turing Award laureate; former VP, Google

    BBC Radio 4, reported by Forbes Secondary

    Cited on The loss-of-control mechanism, step by step

  8. 08

    Summary European Union classifies a general-purpose AI model as posing systemic risk once training compute exceeds 10^25 floating-point operations.

    European Union — Regulation (EU) 2024/1689 (the EU AI Act), Article 51(2)

    EU Artificial Intelligence Act Primary

  9. 09

    Summary Current and former OpenAI and Google DeepMind employees calls on AI companies not to retaliate against employees who publicly raise safety concerns after other channels fail.

    Current and former OpenAI and Google DeepMind employees — "A Right to Warn about Advanced Artificial Intelligence," endorsed by Yoshua Bengio, Geoffrey Hinton, and Stuart Russell

    righttowarn.ai Statement

  10. 10

    Summary Collin Burns found that a GPT-2-level model could help elicit close to GPT-3.5-level performance from GPT-4.

    Collin Burns — Lead author; OpenAI Superalignment team

    "Weak-to-Strong Generalization" (arXiv) Primary

  11. 11

    Summary 28 nations and the European Union commits signatories to AI that is designed, developed, deployed and used safely and responsibly.

    28 nations and the European Union — The Bletchley Declaration, AI Safety Summit

    GOV.UK Statement

    Cited on International treaties and summits: where governance stands

  12. 12

    Summary Simon Lermen undid Llama 2-Chat 70B’s safety training for under $200 using a single GPU.

    Simon Lermen — Lead author, "LoRA Fine-tuning Efficiently Undoes Safety Training in Llama 2-Chat 70B"

    arXiv Primary

  13. 13

    Summary Center for AI Safety holds that mitigating the risk of extinction from AI should be a global priority alongside pandemics and nuclear war.

    Center for AI Safety — Statement on AI Risk, signed by 350+ researchers and executives

    safe.ai Statement

  14. 14

    Summary Dylan Hadfield-Menell formalizes why a rational AI agent resists shutdown unless it remains uncertain about its true objective.

    Dylan Hadfield-Menell — Lead author, "The Off-Switch Game" (UC Berkeley)

    arXiv Primary

Known gaps

This register is incomplete by design rather than by oversight: pages marked as working drafts make claims that are not yet tied to a source here. Those claims are informed judgment, and the page says so. The target is full citation coverage before the manual drops its draft status — progress is tracked in the revision log.

Type to search the manual.

↑↓ navigate ↵ open esc close