Singular State https://singular-state.com Advanced technical expertise for feasibility, prototyping and modernization. We build the working pilot and ramp up your team to take it forward. en-US The EU AI Act's risk tiers, and where a prototype sits https://singular-state.com/articles/eu-ai-act-risk-tiers-and-prototypes/ https://singular-state.com/articles/eu-ai-act-risk-tiers-and-prototypes/ Mon, 07 Sep 2026 00:00:00 +0000 Obligations attach to role, risk class, and the act of placing on the market. Real-world testing is carved out of the pre-market exclusion, and the 2026 amendments moved the high-risk dates. Extend, kill, or transfer https://singular-state.com/articles/extend-kill-or-transfer/ https://singular-state.com/articles/extend-kill-or-transfer/ Mon, 07 Sep 2026 00:00:00 +0000 The decision the operate phase exists to produce. Three named outcomes, three others that get hidden inside them, and a default that applies when the evidence is missing. Kill criteria are a political instrument https://singular-state.com/articles/kill-criteria-are-political/ https://singular-state.com/articles/kill-criteria-are-political/ Mon, 07 Sep 2026 00:00:00 +0000 A precommitted criterion works only where the sponsor survives triggering it. Escalation of commitment is the mechanism it has to defeat, and it was demonstrated in the people who made the original choice. Write the kill criteria before the demo https://singular-state.com/articles/kill-criteria-before-the-demo/ https://singular-state.com/articles/kill-criteria-before-the-demo/ Mon, 07 Sep 2026 00:00:00 +0000 Recorded conditions for stopping, agreed before anyone sees the evidence they govern. Six trigger families, and the rule that a stop decision and a finding about value are separate statements. Model risk management after SR 11-7, and where generative AI sits https://singular-state.com/articles/model-risk-management-generative-ai-scope/ https://singular-state.com/articles/model-risk-management-generative-ai-scope/ Mon, 07 Sep 2026 00:00:00 +0000 SR 26-2 replaced SR 11-7 in April 2026 and its footnote 3 puts generative and agentic AI outside scope. Being out of scope routes the work elsewhere; it does not remove it. NIST AI RMF and ISO/IEC 42001: a framework and a management system https://singular-state.com/articles/nist-ai-rmf-and-iso-42001/ https://singular-state.com/articles/nist-ai-rmf-and-iso-42001/ Mon, 07 Sep 2026 00:00:00 +0000 One is a voluntary risk framework applied to a system, the other a certifiable management system applied to an organisation. Cited as alternatives, they answer different questions and neither certifies an application. When the prototype quietly became a regulated model https://singular-state.com/articles/prototype-quietly-became-regulated-model/ https://singular-state.com/articles/prototype-quietly-became-regulated-model/ Mon, 07 Sep 2026 00:00:00 +0000 The transition is caused by use, not by declaration. A real decision, a real customer, a real record — and the obligations attach retroactively to work already done. Publishing your failure rate is a procurement instrument https://singular-state.com/articles/publishing-your-failure-rate/ https://singular-state.com/articles/publishing-your-failure-rate/ Mon, 07 Sep 2026 00:00:00 +0000 A reported failure rate converts an admission into a screening filter. It only works if the denominator is disclosed, because a zero-failure portfolio is a claim about reporting. Third-party risk management is the gate your prototype actually has to pass https://singular-state.com/articles/third-party-risk-the-real-gate/ https://singular-state.com/articles/third-party-risk-the-real-gate/ Mon, 07 Sep 2026 00:00:00 +0000 A business arrangement can exist with no contract and no payment, so the free pilot is already inside scope. And some obligations survive the relationship ending.