A house newsletter · daily · researched and written by the Kurzia AI
Sunday 15 November 2026 · Issue 17 · 4 min read
Nvidia investors urge Huang's return; AI-built apps leak passports; Italy fines Meta €805M
Today's one thing
Techmeme · yesterday
Nvidia's big investors ask the board: bring Huang back
Takeaway: $2.13 billion
Two of Nvidia's biggest outside investors have written to its board asking it to weigh every option for who runs the company — including a return of Jensen Huang, the founder who resigned as chief executive this week after admitting affairs. CNBC reported the letter.
It turns a settled exit into a live leadership fight at the company the AI industry buys its computing from, and the uncertainty itself is now a planning factor for anyone signing multi-year AI commitments. Boards almost never reverse a conduct-related resignation without paying for it in credibility, so the sensible stance is to plan around the interim chief while the search runs. Brad Gerstner of Altimeter and Gavin Baker of Atreides signed the letter; together they hold $2.13 billion of Nvidia stock.
What is real, what is not
Alpha School — charges parents up to $75,000 a year for a two-hour school day taught by an AI app, and its own figures show most students miss the promised twice-as-fast learning. The software is moving into state schools, which cannot quietly drop pupils who fall behind — a fair test of an AI productivity claim, the same pitch your vendors make. Only 42% hit the goal in both math and reading; roughly half of the younger pupils missed the math target. Fortune
The hottest AI hire is a spy-catcher. AI companies are hiring former intelligence officers to protect trade secrets, and the market rate shows how real they believe industrial espionage is. If frontier labs are spending this, treat your own executives' exposure — conferences, introductions, even dating apps — as a genuine risk to price into your security budget. Counterintelligence salaries at AI firms mostly run $200,000 to $500,000, per Revelio Labs; Anthropic has taken on 33 such staff this year, up from 14 in 2025. Fortune
Money and the market
Mark Stevens — sold roughly $946 million of Nvidia stock across the third quarter, weeks before the affair scandal forced out the chief executive — and filings show he had no pre-set trading plan. Boards should note how fast a conduct story acquires an insider-dealing question when directors trade without one; a scheduled plan is cheap insurance. The sales covered about 4.2 million shares between Aug. 31 and Sept. 18; he still holds about 27.5 million shares. Yahoo Finance
Microsoft Copilot — now runs from a free app to a $99.99-a-month personal plan, while the workplace add-on lists at $30 per user a month paid yearly — with a usage meter on top for advanced features. If a Copilot rollout is on your list, the detail that bites is that only the subscription owner gets the AI on shared family plans, and monthly billing costs 20% more. The business add-on carries an $18-a-seat first-year price for orders placed by 31 December 2026, then $21. gravity.fast
Applied Digital — has been asking tenants including Oracle and Meta for billions of dollars in guarantees, since at least last year, before it will sign their leases — money wanted up front to secure electrical power. AI's bill is shifting from chips to electricity; expect the same pass-through pressure in your own cloud and data contracts. The demands were reported by Bloomberg. Techmeme
Risk the board will raise
Lovable — apps built on the AI platform are leaking private data: a security firm found passport numbers, payment records and over 200,000 chat messages readable with no login. If anyone in your company quickly built a tool on such a platform and pointed it at real customer data, assume it is open until a qualified person has checked who can read it. Crissy Field examined 159,041 projects; Lovable, valued at $13.3 billion, calls it an industry-wide database configuration problem, not a flaw in its AI. Crissy Field
“An easy-to-use development platform does not remove the need for an expert review of access permissions before an application handles confidential data.”
Cursor — confirmed that a key stolen from its AI vendor Fireworks let attackers read stored records of AI coding sessions and then push a malicious copy of Anthropic's coding tool to users — up to 5,000 received it over about 11 hours. Your engineering tools are now a supply chain: a vendor's vendor leaked, and the damage reached customers' machines within days. Cursor
The big AI stories today
Nvidia staff petition — An internal Nvidia post with over 1,600 comments urges the board to bring back Jensen Huang — founder loyalty is now a real force in the leadership fight. WIRED
Meta fine — Italy fined Meta €805 million for reading Android users' photos without clear consent — Europe's privacy regulators will bill any app that handles customer data loosely. POLITICO
Microsoft 365 restored — Microsoft says all Microsoft 365 services are back after an outage that hit governments, traders and hospitals worldwide — check your dependence on one vendor. Techmeme
One takeaway
Crissy Field · yesterday
Ask who can read your AI-built tools
The common thread today is the gap between how easy AI has made building software and how rarely anyone checks it.
Put one question to your team this week: which of our apps — built on an AI platform, bought from a vendor, or both — hold real customer data, and who has actually verified who can read them? Until there is an answer, treat every AI-built tool as public.