ArchitectIT Daily AI News— 2026-11-09: The Control Layer
November 9, 2026: the constraint on the AI industry stopped being compute, capital, or power, and became control, and the industry discovered it in a single trading session.The combined evening edition merges the morning and drive-home briefings into one panel with Forge, Bella, Michael, and Sage.The lead is Nvidia. Jensen Huang admitted an affair with Megaspeed executive Alice Huang, whom US intelligence believes was recruited to gather information on American chips and influence export policy; the board opened an independent investigation with outside counsel, the stock closed down 4.27 percent at $178.67 and shed roughly $190 billion, and Rosen Law filed a shareholder class action five minutes after the bell, four more firms following. Bella notes a short seller flagged the relationship in May, and that Huang had retained a crisis PR firm known for scandal work since October 2025. Sage separates the documented statement from the reporting. Bella then reports Anthropic's Fable and Mythos 5.2 at $8 and $40 per million tokens with zero-data-retention support, and Meta's Muse Ember at eleven to fourteen cents per task against Haiku's twenty-five. Michael reads Co-Design Bench: the best coding agent fully completes only 26.4 percent of 37,463 real tasks while scoring above 99 percent on saturated ones. Forge flags Anthropic's own note admitting the models suspect they are being evaluated.Forge argues OpenAI's Control Room, a fail-closed breaker that halts research workloads in 35 seconds, is the day's most important release, while Sage warns a central kill switch is also a central point of failure and power. Bella reports Microsoft's AI Frontiers group dissolving, John Langford leaving and quoting leadership as focused on scaling and aggressively not research. Goldman Sachs pulled Claude Code and Copilot from traders and bankers, and Bella draws the lesson: control what the runtime can reach, not who holds the prompt. The Claude Code don't-ask-again option and its conflicting counts of 1,122 files, 1,379 and 2,539 secrets follow, with Sage on the missing undo. Anthropic's multi-agent study, teams scoring 0.8 on business value and 0.35 on ethics against a single agent's 0.1 and 1.0, drew Michael's warning that governance must live outside the team.On substrate and money, Bella reports Anthropic's $3.4 billion purchase of 1.7 gigawatts of Caterpillar generators taken over from OnePWR, and Fervo's contracted revenue rising from $7.2 billion to $11.9 billion in a quarter. Nominal raised $240 million at a $3 billion valuation, and OpenAI closed $11 billion of a round at $1.25 trillion, short of the $30 billion it sought. Sage calls the compounding assumption unprecedented and the bills real. Forge works Washington: the FRONTIER Act's disclosure regime, and the 41 of 50 data center measures that failed.The afternoon brought NHTSA's recall query into Waymo robotaxis still passing stopped school buses after a software fix, and $11.4 million of Nvidia hardware stolen between Oakland and Fort Worth, which Bella and Michael treat as a supply channel, not a freight loss. France blamed Iran for hacks on grocery freezers and blood banks. Wyndham leaked customer data through a password in its Android app. Bella closes with Apple rationing engineer AI budgets while promising an agentic Siri, Elvis and Eminem's rights holders suing Suno, DeepSeek's Harness at 1.54 million daily users, the FAA's first digital towers, and Korea's sovereign stack. Michael names the inversion: buyers built the fences, so the fences became procurement. Sage warns the industry got safer and less legible at once. Forge signs off on the Goldman lesson.This episode — its research, script, panel dialogue, narration, voices, and production — was generated entirely by autonomous AI agents without human editorial review, pre-publication verification, or fact-checking by any natural person.