Home / Articles / Underwriting the agent economy, reviewed / Versions / v1.1.0
Underwriting the agent economy, reviewed: what changed in v1.1.0
From v1.0.0 (2026-10-24, 05830a1fe) to v1.1.0 (2026-10-27, db67c5ce1), paragraph by paragraph.
0 paragraphs added, 0 removed, 10 changed in place, 65 unchanged. About 0 words added and 216 removed. Insertions are marked like this, deletions like this; unchanged runs are folded to one line; figures appear as their file names.
2 unchanged paragraphs
!shot uw-cover.webp | images/ | The cover of "Underwriting the Agent Economy: The Blueprint for an AI Insurance Stack", July 2026, arXiv 2607.11999v2, published under CC BY 4.0.2607.11999v2. | screenshot
10 unchanged paragraphs, under In short {#in-short}, The paper in one page {#cliff-notes}
!shot uw-cliff.webp | images/ | The paper on one page: the problem (silent coverage, a disconnect between underwriters and businesses, three providers carrying over 80% of deployments, agent task length doubling every four months),problem, the proposal (affirmative cover in the billions by 2030, an eight-part stack, recurring technical tests, expected-loss pricing),proposal, the precedents (Underwriters Laboratories, the Closed Claims Project, cyber, nuclear) and the tails (a coverage tower up to a government backstop).tails. | concept
3 unchanged paragraphs
!shot uw-failures.webp | images/ | Table 1 of the paper: common failure modes of AI agents (reliability, alignment, security, multi-agent),agents, with the paradigmatic risks and a real example of each. | screenshot
3 unchanged paragraphs, under The AI insurance stack {#the-stack}
!shot uw-stack.webp | images/ | Figure 1 of the paper, the AI insurance stack:stack, with incident data atlooping thethrough topeach feeding standards and accumulation risk research, which feed risk selection and pricing, then ongoing loss control and incident response and claims, which loopstage back to incident data.itself. | screenshot
3 unchanged paragraphs, under Standards, and the precedent that makes the case {#standards}
!shot uw-standards.webp | images/ | Table 3 of the paper: ISO 42001, the NIST AI RMF, STAR for AI and AIUC-1 compared side by issuing body, release, revision cadence, type, focus, whether certification is performance-based, and depth of technical prescription.side. | screenshot
1 unchanged paragraph
!shot uw-ul.webp | images/ | Section II.3.E of the paper: Underwriters Laboratories, founded by property insurers in 1894, testing fire suppression sprinklers around 1919;and the UL mark as a de facto test of insurability, later referenced in electrical codes.mark. | screenshot
2 unchanged paragraphs, under The tails {#the-tails}
!shot uw-tower.webp | images/ | Figure 2 of the paper: a hypothetical coverage tower for frontier AI catastrophic risk, from self-insured retention and primary insurance through mutual insurance, reinsurance and catastrophe bonds to a government backstop.risk. Proportions are illustrative. | screenshot
20 unchanged paragraphs, under The market already selling it {#the-market}, What the paper asks for, and does not build {#the-gap}, What we are building at RiskMandate {#riskmandate}
!shot uw-map.webp | images/ | The paper's eight componentscomponents, on the left; on the right,and what RiskMandate and sgit already produce:produce thefor agenteach. behaviour policy (reach, mandate, the gap), hooks and their audit logs, tests and drills, the delegation record, AIUC-1 as a graph with its conformance layer, insurability as a query, and the maintainers' sign-off on what agents run on. Solid lines exist; dashedDashed lines are proposed. | concept
3 unchanged paragraphs, under What we did with AIUC-1 {#aiuc-1}
!shot uw-insurability.webp | images/ | The AIUC-1 conformance layer on sgit.ai computing insurability for an invented subject on a given date: one condition met, 52 exclusions, none of five consequences covered, each with the reason in the control's own words.subject. Unofficial and derivative; not approved, certified or endorsed by AIUC.derivative. | screenshot
3 unchanged paragraphs, under What the paper has that we still miss {#what-we-miss}
!shot uw-gaps.webp | images/ | What the paper has that we still miss: a loss number, incident data at scale, accumulation, contract language, a recurring test cadence,accumulation and multi-agent failures.more. | concept
15 unchanged paragraphs, under What this does not claim {#limits}, Where this comes from {#sources}