Quaintitative

AIRG, clause by clause

The MAS AIRG: consultation draft vs final

Every paragraph and footnote of the final Guidelines, matched to the November 2025 consultation and compared word by word.

I developed the AIRG while leading AI risk supervision at MAS, but am no longer there, so this is an independent reading. MAS issued the final Guidelines on Artificial Intelligence Risk Management on 7 October 2026, after the November 2025 consultation. Below is the full comparison: every paragraph and footnote of the final text, matched to its source in the consultation and compared word by word. Paragraph numbers follow the final document, with the consultation number alongside. Removed text is struck through; added text is underlined.

For the short version, see what changed and the AIRG is final; for why MAS made these changes, the response to feedback; for the guidelines themselves, the AIRG explained.

Key changes

    removedadded
    Method: text was extracted from both PDFs and matched paragraph by paragraph, then compared word by word. Draft list markers (a., b.) were normalised to the final style ((a), (b)) so relabelling does not show as a change. Superscript footnote markers are omitted from body text; footnotes are compared in their own section. Section headings, the cover, contents page and the two figures are not part of the comparison.
    Status labels are an editorial judgement: Substantive changes alter an expectation, scope or threshold; Editorial changes are wording or drafting only.

    Get more

    I am publishing more details on AI risk management in the coming weeks. Subscribe to get them, and future updates on the AIRG.

    Subscribe for updates