for agents/llms.txtv0.13.110

Home / Articles / Four frameworks, one agent / Versions / v1.2.0

Four frameworks, one agent: what changed in v1.2.0

From v1.1.0 (2026-10-13, f5462b64c) to v1.2.0 (2026-10-27, db67c5ce1), paragraph by paragraph.

0 paragraphs added, 0 removed, 4 changed in place, 62 unchanged. About 0 words added and 204 removed. Insertions are marked like this, deletions like this; unchanged runs are folded to one line; figures appear as their file names.

← v1.1.0 · all versions

2 unchanged paragraphs

!shot ff-map.webp | images/ | Four frameworks, as iason's paper reads them, against what runs here.here, Formarked each:as the question it asks of a system, how the paper summarises it, what answers it on this site and at RiskMandate, the article or vault where that can be checked, and whether it runs, isrunning, designed or is only argued. Green runs; amber is a design; grey is an argument. | comparison

15 unchanged paragraphs, under In short {#in-short}, The paper, and what it gets right {#paper}, The unit: a system before deployment, or one agent in one deployment {#unit}

!shot ff-unit.webp | images/ | Two units. On the left,units: the frameworks' unit: an AI system, described by its characteristics, context and impact, given a tier, and assigned obligations, before it is deployed. On the right, the unit here:here, one agent in one deployment, with its reach measured, its mandate written, the gap between them typed by what stands in the way, its footprint read afterwards, and a blast radius for every row. The arrows show where each of the four frameworksframework attaches. | architecture

16 unchanged paragraphs, under OECD: five dimensions, and the two that matter most for an agent {#oecd}, NIST: govern, map, measure, manage, as they run here {#nist}, Japan: three roles, one policy between them {#japan}…

!shot ff-dates.webp | images/ | A status since, forsince the paper.paper: Above the line, theits dates as the paper gives them, as of November 2025:2025 prohibitions in force from February 2025, and the rules applying in full by August 2026. Belowabove the line, and what this site's own articlessite recorded afterwards:afterwards the amending regulation in force from 27 July 2026, Annex III obligations from 2 December 2027, Annex I from 2 August 2028, and software a product under the EU's new liability rules from 9 December 2026. Neither row is wrong on its day.below. | timeline

4 unchanged paragraphs, under The five principles on one small system {#principles}

!shot ff-principles.webp | images/ | The paper's five shared principles, each against what answers it here, the evidence, and how far it goes: runs, in green; a published design with a demo behind it, in amber. Four of the five run, on one newsroom and one agent team; risk acceptance as a record is the design.goes. | comparison

25 unchanged paragraphs, under What the graphs add, and the one to build next {#graphs}, Where it is hardest {#hardest}, What this does not claim {#limits}…

← v1.1.0 · all versions