AiR Bench

Retrieved 2026-09-22. DATE FIELD = retrieval date. Dating evidence actually retrieved: the page text states 'v1.0 of the Econ Scenario Explorer, September 2026' and the search engine reported page age of 5 days, so publication appears to fall inside the six-week window; exact day not verified, so not asserted. Content: a publisher-built scenario model of AI's macroeconomic paths, with an explicit self-stated disclaimer that it isolates a few forces and omits policy responses, business cycles, aggregate demand or financial-market disruption and catastrophic risks. Maps to R5 (frontier response) as a primary-publisher artefact showing how a frontier lab frames economic transition, and weakly to D1 (strategy scenario planning). IMPORTANT LIMIT: this is a model/scenario tool from an interested vendor, not measurement of firm outcomes; it contains no UK, mid-market or realised-value data and must not be used as evidence of business value. Logged as informing context only.

Retrieved 2026-09-22. DATE FIELD = retrieval date, not publication date: the search engine reported page age of ~1 month (so roughly mid-to-late August 2026, probably inside the six-week window) and the page itself carried no explicit publication date I could read, so I will not assert an exact publication day. Finding: 37% of respondents say AI has contributed positively to their organisation's EBIT, 'essentially unchanged from 2025' despite a rising share of organisations scaling AI; non-EBIT gains are reported in innovation, differentiation, customer and employee satisfaction, and function-level cost reduction is most often reported in supply chain and service operations. Maps to R3 (value realised) and R2 (conversion from adoption to measured P&L effect): scaling is decoupling from enterprise-level financial impact, i.e. deployment is growing faster than conversion. Also touches R4 (value concentration) indirectly, since enterprise-level EBIT impact stays with a minority. LIMITS for UK mid-market use: global, cross-size respondent panel; no UK or mid-market cut visible in retrieved text; self-reported perception, not audited financials. Treat as directional context, not UK mid-market evidence.

Retrieved 2026-09-22. DATE FIELD = retrieval date. No publication date was available: the search engine returned no page age and the retrieved text carried no date, so publication may well pre-date the six-week window; flagged as watching rather than entered for that reason, and no publication date is asserted. Finding as retrieved: average responsible-AI maturity score rose to 2.3 in 2026 from 2.0 in 2025, but only about one-third of organisations reach maturity level 3+ in strategy, governance and agentic-AI governance; survey fielded December 2025 to January 2026 with roughly 500 organisations across industries and regions, respondents holding AI governance, risk or investment responsibility, scored against McKinsey's five-dimension AI Trust Maturity Model. Maps to D5 (governance immaturity lagging deployment), D1 (strategy) and R6 (agent governance specifically weakest). LIMITS: fieldwork predates the window; small self-selected panel; vendor maturity model with no published scoring detail retrieved; no UK or mid-market cut. Action: verify publication date and look for a UK cut before using.

Out-of-window context, logged as informing only: publication date 2026-07-20 per the search result, which is outside the requested six-week window (2026-08-11 to 2026-09-22). Only the title, URL and publication date were retrieved via search; the article body was NOT read before the search budget was exhausted, so no substantive statistic is recorded here and nothing is asserted about its contents. Relevance if read: ONS is the primary official source for UK business AI adoption incidence and would bear mainly on R1 production and, depending on question wording, D4 data and D6 execution. Flagged for follow-up in the next research cycle to check (a) whether a newer ONS release or Business Insights and Conditions Survey module has appeared inside the six-week window, and (b) whether size-band breakdowns permit a genuine mid-market (rather than SME-versus-large) cut.

OpenAI shipped ChatGPT Work, an enterprise agentic service with GPT-5.6 Sol as its operating agent, plus GPT-Live voice models, moving flagship capability from model to deployed agentic product. A developing signal on real agentic deployment and pilot-to-production paths; informs R6 (agentic deployment), R2 (pilot-to-production conversion) and D3 (process readiness). Watching pending independent evidence of mid-market uptake and value.

OpenAI publicly launched the GPT-5.6 family in three tiers (Luna, Terra, Sol) positioned for enterprise work, agentic coding, scientific research and cybersecurity, after a limited 26 June preview to 'trusted partners' imposed under US government restriction. A material frontier step within the window; informs R5 (frontier responsiveness), D6 (execution track record) and D1 (strategic clarity as the reference bar mid-market firms benchmark against).