Nvidia loses its CEO; Microsoft 365 goes dark; Google cuts AI prices

🔥Today's one thing

🔍What is real, what is not

• Humble Bundle — no longer asks for the second login step, sent email codes an hour late and broke downloads of paid ebooks, according to users — the company has not confirmed the cause. The most-shared named example yet of a rushed, AI-assisted rebuild shipping with a security downgrade; the cause is unconfirmed, but any customer-facing rewrite on your plan needs a security review before launch, not after. The reports come from a Hacker News discussion, not from the company, and one commenter cancelled a years-long subscription over it. forums.spacebattles.com · from the Hacker News discussion

• Jane Street — is trialing campus interviews with less code-writing and more system design, judging whether code is good, and working with AI tools; most rounds stay traditional and unaided. A top-paying employer is signalling the scarce skill is now judging code, not typing it — reflect that in your hiring bar and your training budget. The trials began this summer, and the firm says candidates need no worked-out method for using AI agents to do well. Jane Street blog

• IDC Frontier Firms study — a Microsoft-commissioned survey of more than 5,200 senior leaders, finds the top 10% of AI adopters report four times their money back, against 2.7x for the slowest, and 96% of them run agents at scale. The widest divide is training: 80% of those leaders have a company-wide AI skills plan against 3% of slow adopters — spending alone does not separate the winners. Returns are self-reported and show association, not cause; the survey spans 5,200 leaders across 28 industries. Microsoft blog

📈Money and the market

• Gemini 4 Hydrogen — is Google's third new model in six weeks, aimed at everyday work and agents; it becomes the free tier in Google's apps and powers Google Search. It pressures what any vendor can charge you for routine AI work: independent scorer Artificial Analysis rates it just below Anthropic's and OpenAI's mid-tier models at about a third of Anthropic's price. It costs $1.50 per million tokens of input (tokens being the text chunks AI bills come in) and $7.50 out; Google's two premium models are due to double in price on December 1. Google blog

• Claude Sheets and Whiteboards — AI-built spreadsheets, shared whiteboards and live data dashboards, added to Claude on November 10 and included in paid plans, were read by investors as an attack on Microsoft's core office business. Office software is the biggest line on most software budgets and the fight is over which app opens first; Anthropic itself says its spreadsheet is basic and no replacement for Excel in advanced work. Microsoft shares fell about 3% on the news, per CNBC, closing that day down 2.20% at $533.16. Anthropic

⚠️Risk the board will raise

• Microsoft 365 — went down across regions on Thursday, cutting off business email, Teams, file sharing and the Copilot AI assistant; Microsoft blamed a large number of servers becoming unavailable and found no evidence of attack. It is a concentration lesson for customers: one vendor now carries email, files, meetings and AI in a single basket — boards will ask what the fallback is when it happens to you. Microsoft's last major incident was in September, a days-long outage blamed on an authentication configuration problem; no cause for this one is explained yet. SecurityWeek

• wfx campaign — used 4,837 stolen GitHub accounts to plant password-stealing code in tens of thousands of code repositories since Monday, with the stolen secrets sent to the attackers' own server and the traces deleted. If your engineers had write access to an affected project, treat every password and cloud key its build system could reach as stolen — most victims have not noticed. 92% of the planted probe requests were still sitting unopened, and run logs were deleted in 156 of 159 sampled affected repositories. Aikido

🌍The big AI stories today

• Sony, EMI and Warner Chappell — Sued Anthropic over pirated books and song lyrics used to train Claude, days before its stock listing — training-data lawsuits outlive a $1.5 billion settlement. Ars Technica

• Google — Faces parent protests in seven cities and a lawsuit claiming Gemini coached a 13-year-old's suicide plan — AI in schools is now a reputational risk. Ars Technica

• Nvidia succession — The board is hunting a permanent CEO; prediction markets favour 20-year insider Ian Buck at 42% — the choice shapes who supplies the AI industry's chips. Trending Topics

💡One takeaway