Humanity & AI Safety Accord
Menu

The Accord

Version history

Once a version is published it never changes. If the wording changes, a new version is created and the old one stays exactly as it was, so every signature keeps pointing at the precise text that person agreed to.

  1. Version 1.1

    publishedSeptember 22, 2026

    Revised wording: benefits for every nation as an aim, a meaningful public voice, technology companies at the table, and humanity keeping meaningful control over increasingly capable systems. The summary now explains the Accord rather than committing the signer. Wording is pending domain-expert and legal review.

    Hash
    68c352635deac3da302cbdcc26d96c17f370c611180e712b2de06ad81ed60821
    Approved by
    Pending governance council
  2. Version 1.0

    retiredSeptember 13, 2026

    Repositioned per change spec v1.1. Materially different from v0.1: the five named principles, including slowing frontier development, are replaced by a broad, non-prescriptive request for government oversight and international cooperation. Signatories to v0.1 agreed to different words and are not counted under this version.

    Hash
    ba7352adbf367a3c49472cc399e422839d3b4ae66c9328a18f0be13a4ece9391
    Approved by
    Pending governance council
  3. Version 0.1

    retiredSeptember 13, 2026

    First published draft. Wording is pending domain-expert and legal review.

    Hash
    5800f7619e09461cf291aaf9a3a3e907826dfe29deb3f8a3ae4ea988b3589f4e
    Approved by
    Pending governance council

Read the current Accord