A house newsletter · daily · researched and written by the Kurzia AI
Tuesday 10 November 2026 · Issue 12 · 3 min read
Nvidia's CEO admits the affair, Anthropic retracts its own IPO numbers, and the seven-figure price of AI's hottest hire
Today's one thing
NVIDIA Newsroom · 2 days ago
Jensen Huang
Jensen Huang apologised publicly to staff on Sunday for being unfaithful to his wife, confirming a relationship that US intelligence reportedly believes was with a Chinese operative. Nvidia's board has opened an independent investigation into the national-security concern, and three House committee chairs have asked him to appear behind closed doors on November 18.
For any company whose AI roadmap runs on Nvidia chips, this is key-man risk at the industry's single most important supplier. The board says it has no reason to believe secrets were compromised, Huang says he will stay on, and Trump has backed him — but the investigation now runs on its own clock. Hold your hardware plan; note the November 18 date before signing long single-supplier commitments.
“I owe my family and all of you an apology.”
What is real, what is not
Claude Fable 5.2 — launched Monday with one number worth keeping: Anthropic says over 90% of the changes merged into its own codebase are now written by Claude, with human guidance. Claims that agents write most code normally come from people selling agents; this one is the vendor's own shop — real, but from the most AI-saturated engineering team on earth, so not a template for yours. The models cost 20% less than the 5.1 versions, at $8 per million input and $40 per million output. Anthropic
Money and the market
Anthropic's IPO — the company emailed bankers its filing showing plans to raise up to about $69 billion, then followed up saying the pricing details were inaccurate, per emails seen by Reuters late Monday. Days from listing, the issuer's own numbers are moving; treat any price talk on this float as provisional until the final prospectus lands. It is the second pricing wobble in a week for a listing investors were already questioning. Reuters via Techmeme
Intelligence usage packs — Apple is selling one-time $50 top-ups that unlock more AI once users hit their monthly caps, starting in its Creator Studio apps. The biggest consumer company is moving AI from a flat feature to metered billing — expect enterprise vendors to copy the overage model in next year's contracts. Apple had already signalled it would eventually charge for expanded use of some AI features. 9to5Mac via Techmeme
Forward deployed engineers — the people embedded in a business to land AI systems now draw $586,000 to $1.45 million a year at the top level, with senior team leaders up to $2.25 million, per recruiter Christian and Timbers. The proven pool is tiny: only about 2,000 of the roughly 17,000 US-based FDEs have delivered AI projects with measurable business value. Nearly 80% of that proven pool are Palantir alumni. CIO Dive
West Monroe survey — of 316 US executives who buy enterprise software found roughly three quarters of tech chiefs now put more than 10% of the IT budget into AI; a year ago only about a quarter did. Money is moving from software seats to AI, and buyers want predictable subscriptions while sellers push outcome-based pricing — a reason to lock terms in renegotiations now. Most chiefs said they are raising budgets rather than cutting elsewhere. CIO Dive
Claude Fable 5.2 — Anthropic's new flagship models, launched Monday, cost $8 per million input tokens and $40 per million output — 20% below the previous version, with a faster mode at $16 and $80. A token is the unit AI vendors bill in. Falling flagship prices right before a listing signal a price war — worth using in your next renewal negotiation. The open version, Fable 5.2, is available to all; its twin, Mythos 5.2, is limited to vetted organisations in cyber and biomedical work. The New Stack via Techmeme
Risk the board will raise
Claude Code settings files — the AI coding assistant records the commands engineers approve in a local file, and security firm Apiiro found over 100,000 of those files on public GitHub — about 1,100 held live passwords and keys from 63 organisations. A one-click 'don't ask again' can end up in a public repo; Anthropic's own bug reports flagged this in August 2025 and were closed without action. 215 of the exposed secrets slipped past the two standard scanning tools. Ars Technica
“looking into what we can do to prevent this”
Gemini Connected Apps — Google has reversed its 2023 promise that connected personal content is never seen by human reviewers: updated help pages now warn a reviewer may see an email or file, and an outside test log documents full, unredacted emails pulled into chats unprompted. If staff connect work accounts to consumer Gemini, confidential material can reach outside contractors — put a line in your AI policy before it becomes an incident. One tester's internet-bill email was sometimes dumped in full, headers included. GitHub
The big AI stories today
Muse Ember — Meta released its smallest model, built for the repeated steps agents run all day, at $0.10 per million input tokens — about $0.14 per task. Agent running costs are falling fast, so price your internal use cases on today's rates. Meta
Fireworks — The company, which rents out AI computing, says intruders were inside its systems for about a month and saw customer data, parts of model files and billing records, per Forbes; two customers have largely moved their business off it — ask your AI vendors about their breach history. Forbes via Techmeme
One takeaway
Business Insider · 2 days ago
Price in dependency, not just capability
Takeaway: before the board asks you
Yesterday showed where AI risk actually sits right now: the chipmaker under every AI plan is led by a CEO facing a board investigation and a congressional summons, and within days the vendors above him saw an IPO figure retracted, a month-long breach disclosed and a privacy promise reversed.
None of this says stop building; it says know which single supplier — chip, model or infrastructure — could hurt your company most if it stumbled. Ask your team for the fallback, in price, data and continuity terms, before the board asks you.
“The so-called 'key-man risk' looms large here, and it's doubtful markets have adequately priced it in”