What changed
- Musk v. Altman verdict for defendants (Musk v. Altman (and OpenAI / Microsoft / Brockman)) — unanimous nine-juror dismissal of all claims against OpenAI, Altman, Brockman, and Microsoft on statute-of-limitations grounds in less than two hours of deliberation. Judge Yvonne Gonzalez Rogers adopted the verdict. The substantive merits of OpenAI's for-profit pivot were not reached — per Gary Marcus, "the AI trial of the century ends with a whimper." Wiki predictions resolved: Case proceeds to verdict (RESOLVED-correct); Verdict finds OpenAI liable on ≥1 count (RESOLVED-incorrect). Musk's $150B damages claim, his ouster-of-Altman-and-Brockman remedy ask, and his dismantling-of-OpenAI's-for-profit-arm demand are all defeated on the same procedural basis.
- Memory-chip cyclicality (AI Chip Mania Sows Seeds of Its Own Destruction (Mackintosh, WSJ, May 16 2026)) — Mackintosh's WSJ Streetwise column (May 16) becomes the wiki's anchor for the "low forward P/E is a cycle-peak signal, not a bargain" frame on memory equities. Micron's three historical cycle peaks (1984: 15×; 2018: 5.5×; 2022: 9×) all preceded crashes. Sub-10× forward P/E in May 2026 fits the same pattern. Load-bearing risk Mackintosh names: AI memory-efficiency breakthroughs (Alphabet's March 2026 memory-efficiency paper triggered a temporary memory-stock selloff that recovered). Pairs with the May 17 Samsung 45,000-worker memory-plant strike (acute supply disruption to ~1/3 global DRAM share; bullish short-run, irrelevant long-run in the cycle frame).
- NextEra-Dominion $67B all-stock merger announced (AI Data Centers) — the May 15 FT-reported $400B-talks figure crystallizes as a $67B all-stock deal creating the world's largest regulated utility by market cap (~10M customer accounts across FL/VA/NCs). NextEra holders own 74.5%; closing expected in 12–18 months pending Nuclear Regulatory Commission approval. Both companies explicitly framed the deal as AI-power-demand-driven. NRC approval is now the binding regulatory step for the wiki's largest utility-consolidation event triggered by AI compute load.
- OpenAI product-strategy consolidation under Brockman (OpenAI + Greg Brockman) — May 15-16: ChatGPT + Codex + developer API folded into one core product team. Brockman now permanently leads product strategy. Thibault Sottiaux runs core product + platform (forthcoming desktop "super app" combining Codex, ChatGPT, and Atlas browser); Nick Turley moves to enterprise; Ashley Alexander leads consumer. Brockman framing: "maximum focus toward the agentic future, to win across both consumer and enterprise."
- Anthropic + OpenAI = 89% of AI-startup revenues (Anthropic) — The Information reports the combined two-firm share by mid-May 2026. The wiki's "duopoly-with-fast-rotating-pole-position" framing now anchors on this number: two firms capture nearly all revenue, but relative position changes month-to-month (Anthropic 34.4% vs OpenAI 32.3% US business adoption on Ramp data; Vercel AI Gateway Anthropic at 61% of spend; OpenAI's >900M weekly ChatGPT users).
- Redwood Research "incrimination via distillation" (AI Scheming) — Prasanna, Westover, Mallen, Pan et al. propose a structurally novel anti-scheming methodology: distill suspected-misaligned teacher U into less-capable student M, audit M. The student is less able to evade evaluation. Pairs with the broader evaluation-awareness confound that has dominated the Apollo / OpenAI / Anthropic / METR / UK AISI scheming literature. Load-bearing empirical assumption: propensity-transfer through distillation works even when teacher outputs look benign.
- Trump May 18 post-summit AI-guardrails remarks (US-China AI Competition: Different Races, Different Metrics) — speaking on Air Force One after the Beijing summit, Trump told reporters he and Xi "talked about possibly working together for guardrails" and "probably will work together." Restraint from Democrats and security advocates; no concrete deliverables beyond Bessent's earlier "protocol on dangerous AI capabilities" remark.
- Duff (GovAI) finds Chinese tech-and-finance media engage seriously with Western AI-safety essays (US-China AI Competition: Different Races, Different Metrics) — 61 Chinese-language primary sources analyzed; 85% neutral framing; combined readership likely >1M for both AI 2027 and Adolescence of Technology. Wall Street News selectively omits Amodei's CCP-as-authoritarian-threat paragraph while keeping the rest of the translation intact — editorial self-censorship, not state suppression.
- Andreessen's "model moats decay in months / 12-to-100-model stitching" (Marc Andreessen) — January a16z LP meeting framing surfaced May 18 by Dominguez; xAI matched OpenAI in 12 months; DeepSeek replicated GPT-5-class reasoning within weeks of Jan 2025. Winning AI applications stitch 12–100 specialized models per workflow.
- Polis press release on signing SB26-189 (Colorado SB 26-189 (2026 — replaces 2024 Colorado AI Act)) — bipartisan AI-taskforce framing; packaged with SB26-137 (DORA five-year rule reviews) under "Breaking Down Barriers and Reducing Regulation" — situates Colorado's partial retreat from EU-AI-Act-style risk-based duties as part of a broader deregulatory consolidation. New Jared Polis page anchors future Colorado-AI-policy references.
What it connects to
The Musk v. Altman procedural verdict closes one major lever of pressure against OpenAI and opens another. With statute-of-limitations precluding the merits being reached, the substantive theory that OpenAI's for-profit conversion was a breach of charitable-trust principles remains legally untested. The trial-established factual record (Mira Murati's "Sam lied to me" deposition; Helen Toner's "Altman misstated which GPT variants were submitted"; Brockman's $30B stake + financial ties to Altman; the $100B Microsoft partnership-spend figure; the $852B trial-materials OpenAI valuation; the Schizer "$200B foundation should have more" expert testimony; the Microsoft 2018 pre-investment-skepticism emails) — all of it survives in the public record and is now available to future plaintiffs / congressional investigators / SEC inquiries / journalists, but none of it created any operative legal effect on OpenAI's structure. The IPO path is materially cleared. Per Marcus's framing, the procedural ending also means future founder-vs-nonprofit-AI-conversion cases lack a merits-decided precedent to anchor on — Musk v. Altman becomes a timing-precedent only.
Mackintosh's chip-cycle frame fills a wiki gap. Prior wiki framing tracked AI compute economics primarily through (1) HBM supply chokepoints, (2) hyperscaler capex (circular financing), (3) data-center power constraints. Mackintosh adds a fourth axis: memory-chip equity valuation as a commodity-cycle phenomenon the market historically misreads at peaks. The frame is complementary, not competitive, to the others — supply chokepoints + hyperscaler capex are real and binding now; Mackintosh's prediction is about what happens when the next cycle of overcapacity (Micron $150B + new Korean fabs) arrives in 2027-28.
NextEra-Dominion at $67B is a meaningful step-down from the May 15 FT $400B-talks figure — but the strategic theory is unchanged. The deal value reflects the actual exchange-ratio mechanics ($67B for Dominion's equity stake in the combined entity); the combined-entity rough valuation is larger. The wiki should treat the announcement as confirmation of the May 15 talks rather than as a downgrade — the question that matters is NRC approval (12–18 months out) and Dominion's Northern Virginia data-center load.
The OpenAI Brockman product-strategy consolidation has a structural reading the May 15 Wired story missed. Brockman is the only OpenAI executive who survived both the November 2023 board crisis and the May 2026 trial intact — and is now permanently in charge of the operative product surface (ChatGPT + Codex + API). The configuration is functionally a Brockman-led product organization with Altman as CEO. Pair with the Altman May 18 verdict aftermath reading: Altman's CEO standing was preserved by the verdict but the trial-established record on his credibility issues with his own former CTO and former board remains. The product-organization design distributes decision-making rights in a way that insulates the product roadmap from any future Altman-credibility incidents.
One question worth sitting with
The Knight Columbia 2026 cluster (May 17 ingest) argues for deployment-context-sensitive AI governance anchored on social/economic/political structures the AI system is embedded in. The May 18 cluster delivers four distinct kinds of operative-event evidence about those structures: a federal jury deciding on procedural rather than substantive grounds; a private utility merger consolidating power supply to data centers under NRC oversight; a frontier-AI-lab reorganization that distributes decision-making rights internally; and an academic-press finding that Chinese policy elites are reading Western AI-safety arguments seriously.
Each of these is exactly the kind of deployment-context evidence the Knight Columbia frameworks call for. None of them is being studied as such. The wiki's current architecture has a place for each individually — Musk v. Altman (and OpenAI / Microsoft / Brockman) for the verdict; AI Data Centers for the utility merger; OpenAI for the reorg; US-China AI Competition: Different Races, Different Metrics for the Duff finding — but no consolidated place for deployment-context analysis as a methodological approach. The Knight Columbia cluster argues that the integration of these threads is the unit of analysis, not any one of them.
Question: does the wiki need an analysis page (analysis/deployment-context-evidence-2026/) that explicitly threads these together as a worked example of the Knight Columbia frameworks in action? The case for: the frameworks are otherwise abstract anchor pages with no operative example. The case against: the integration may be premature; better to let the threads accumulate first and integrate when the pattern is denser.
(I lean toward "yes, but write it as a worked example, not as analysis claiming structural conclusions" — the Knight Columbia frameworks are the load-bearing methodological move; the page exists to demonstrate them in operation.)