the week in AI, briefly. then briefly again.
Artificial Intelligence — briefly, then briefly again · tbb.ceo
listen tothe weekly #028 0:00 –:––
#028 ·13 NOV 2026 ·FRIDAY ·5 MIN READ ·10 STORIES + 8 EXTRAS

The Board, the Benchmark, and the Ballot Box

Three forces shaped AI this week: a governance crisis at the world's most valuable chip company, a measurement proving models now accelerate the labs that build them, and 40 ballot measures telling data centres to build somewhere else. The bottom line: the industry's constraints have shifted from compute to trust — in institutions, in valuations, and in the communities asked to host the infrastructure.

01 / The Ten

The week, ranked

10

Nvidia board moves to oust CEO Jensen Huang amid spy scandal

Nvidia's board is preparing to remove its co-founder and 32-year CEO following admitted affairs — one with a woman US intelligence identifies as a foreign agent — triggering congressional hearings, an employee petition for his return, and the surreal spectacle of Trump floating Tim Cook as successor. CFO Colette Kress is discussed as interim replacement.

Why it mattersThe founder-CEO of the world's most critical AI chip company may be gone within days — a governance crisis with direct implications for AI supply-chain continuity, US-China compute policy, and the lobbying apparatus that kept export controls from tightening faster.

METR audit: Anthropic may have crossed its AI R&D acceleration threshold

METR evaluated Anthropic's latest internal models on realistic R&D tasks and found they completed 25–50% more research sub-tasks per hour than human baselines — the first formal documentation that a lab's models are measurably speeding up its own development cycle. Acceleration concentrated in experiment design and literature synthesis.

Why it mattersIf models can accelerate the lab building them, the feedback loop becomes self-reinforcing. This is the measurement the safety community said would matter — now it exists.

Anthropic launches Claude Fable 5.2 and Mythos 5.2

Anthropic released its fastest model family: Fable 5.2 scored 65 on Artificial Analysis — highest of any model — while Mythos 5.2 is gated for vetted security and biosecurity researchers. Fast mode reaches 2.5x speedup over Fable 5.1; pricing unchanged.

Why it mattersThe benchmark result sets the floor for the next round of model releases from competing labs. Also demonstrated: Mythos compressed a multi-month AMD GPU migration to roughly three days with minimal human engineering.

OpenAI closes $11B at $1.25T, surpassing Big Tech peers

OpenAI raised $11 billion at a $1.25 trillion valuation — ahead of Meta and trailing only Apple, Microsoft, and Nvidia among US tech firms. Gulf sovereign wealth funds contributed a reported $4B slice. Capital earmarked for compute infrastructure and Stargate Phase 2.

Why it mattersThe number reframes the competitive floor. Any rival model lab now justifies its valuation against a $1.25T comparator — Anthropic's IPO pricing conversation just got harder.

DeepMind's AlphaProof Code verifies Signal crypto at machine level

Google DeepMind released AlphaProof Code, an agentic framework that used Gemini-Carbon-powered prover agents to mathematically verify the Signal protocol's reference implementation down to raw AArch64 machine code — generating 6.5 million lines of Lean proofs in four weeks, a task that previously took human verification teams years.

Why it mattersAutomated machine-level formal verification, if reproducible at scale, changes the assurance ceiling for critical software — a genuine role for AI agents beyond chat.

Google launches Gemini Decisions API: 1,000 questions, one pass

Google DeepMind released a model built not for generation but for scoring. Gemini Decision 1 evaluates up to 1,000 typed questions against multimodal inputs in a single forward pass, returning calibrated probabilities at $0.10 per million input tokens with no output charges.

Why it mattersA purpose-built decision model at this price point changes how developers build agentic pipelines — cost-of-inference for orchestration just dropped by an order of magnitude.

OpenAI quarterly revenue crosses $10B for the first time

Q3 revenue reached approximately $10 billion — a roughly 50% jump from $6.7B in Q2 — driven by enterprise API demand. Operating losses reportedly widened despite the surge. Anthropic is achieving higher revenue with better operating efficiency.

Why it mattersThe milestone sets a credible revenue baseline for OpenAI's forthcoming IPO while surfacing that Anthropic may be the more efficient operator at scale.

Anthropic's IPO faces pushback; accidental pricing leak compounds

Institutional investors pushed back on Anthropic's $1.5T IPO target, with large funds signalling a walk-away price below $1.4T. Days later, Anthropic distributed an IPO document to underwriters containing pricing information it subsequently disowned — a gaffe exposing coordination problems in the pre-IPO process.

Why it mattersThe gap between a founder's target and what buyside funds will pay is where listings stall. The pricing leak suggests internal disorganisation or a rushed timeline, neither of which helps institutional confidence.

Voters rejected 40 of 41 data centre ballot measures

Post-midterm analysis found that local zoning proposals to fast-track AI data centre construction were defeated in 40 of 41 US jurisdictions. Opposition was bipartisan and locally driven — no national campaign coordinated the rejections. Affected areas include grid-connected corridors in Texas, Virginia, and the Mountain West.

Why it mattersLocal democracy just became the fastest-emerging bottleneck in AI infrastructure — faster than energy permitting or chip supply.

Anthropic ships Claude Sheets, Whiteboards, and third-party connectors

Anthropic launched Claude Sheets (spreadsheets with formula support and .xlsx export), Whiteboards in beta, and graduated Claude Data to GA — alongside interactive third-party connectors that run Atlassian, Adobe, and HubSpot tools in-pane. A direct challenge to Microsoft 365 Copilot.

Why it mattersAnthropic moving from chat into collaborative workspace puts it in direct competition with productivity suites for the first time — a significant strategic expansion beyond model-as-API.
02 / Also

Worth knowing

08
Google launches Gemini 4 Hydrogen
The lowest-cost Gemini 4 family member, a Flash-line replacement, ships first to Google Search worldwide. Early benchmarks put it close to Claude Sonnet 5.5 on Artificial Analysis while costing less per token.
deepmind.google ↗
World Labs ships Frameboy: real-time 3D worlds in browser
Single image to explorable 3D scene in under two seconds at 60fps, running via WebGPU with no local GPU. Free tier live; API opens in 30 days.
worldlabs.ai ↗
Anthropic gives Claude live bank account reading
Claude gained read-only bank access via Plaid — categorising 90 days of transactions in conversation. US only; UK/EU planned for Q1 2027.
the-decoder.com ↗
Corporate AI spending stalled in October
Ramp data showed AI tool subscriptions growing at their slowest rate since early 2024. Copilot cancellations outpaced new activations for a second month. Inference API spend, however, grew 14%.
techcrunch.com ↗
OpenAI deploys Control Room kill-switch for model workloads
A centralised infrastructure layer that monitors and can terminate errant model workloads — including live agentic inference — within 35 seconds. Fail-safe: if Control Room disconnects, generation halts.
openai.com ↗
Frontier Model Forum releases CVSS-style jailbreak rating standard
The Anthropic/OpenAI/Google/Microsoft consortium published a four-vector scoring system standardising vulnerability disclosure for frontier LLMs. AI security has moved from ad-hoc disclosure to standardised scoring.
frontiermodelforum.org ↗
Chinese AI firms rush overseas Nvidia compute before rules close
Firms securing GPU capacity in Singapore, Malaysia, and UAE data centres as Commerce finalises rules blocking remote access to US AI compute by Chinese entities.
reuters.com ↗
Anthropic: Mythos ported Claude to AMD GPUs in 3 days
In its IPO roadshow, Anthropic disclosed Mythos compressed a multi-month AMD GPU migration to roughly three days. Claude now runs on four chip families: Nvidia, AMD, Trainium, and TPUs.
the-decoder.com ↗
Be subscriber #028 weekly AI email interest list
Back issues

The Archive

Register interest

Get the brief

weekly AI email interest list. Read the latest briefs on the site. 27 signups recorded.