[AI WEEKLY NEWS RUNDOWN] The Week AI Found Its Single Points of Failure (Nov 15, 2026)

🎧 Listen ad-free:


🌐 Daily Briefing: https://djamgamind.com/daily
📚 Research Hub: https://djamgamind.com/pdfs

🔍 Keywords: AI Weekly News, Nvidia, Jensen Huang, Colette Kress, Microsoft 365 Outage, Claude Fable 5.2, Gemini 4 Hydrogen, Muse Ember, GPT-6 Cyber, ChatGPT Wallet, Claude Sheets, Grok Build, AI Agents, AI Resilience

This week’s question: What is your single point of failure?

Summary:
On Monday, Nvidia’s board opened an investigation into its CEO. By Thursday, Jensen Huang had resigned after 33 years, and CFO Colette Kress was running the company. The same week, a Microsoft 365 outage reached 74 countries and cut ChatGPT and Claude off from Outlook, and two security reports traced thousands of leaked secrets and 35,578 open apps to default settings. Meanwhile the labs kept shipping: Anthropic’s Fable 5.2, Google’s Gemini 4 Hydrogen, Meta’s Muse Ember, OpenAI’s GPT-6 Cyber and a wallet for agents.

Important Topics:

  • One person: Nvidia loses its founder in four days.

  • One supplier: the Microsoft outage, and Nvidia’s $530 billion checkbook.

  • One default: leaked secrets, open apps, and agents that buy anyway.

  • New models: Fable 5.2, Gemini 4 Carbon and Hydrogen, Muse Ember, GPT-6 Cyber.

  • New products: ChatGPT Wallet, Claude Sheets and Meetings, Businesses in ChatGPT, Gemini glasses.

  • The money: OpenAI’s quarter and Anthropic’s IPO.

  • This week’s corners: Cert Corner, Red Flag or Green Light?, and the Boardroom Call.


👤 One person: Nvidia loses its founder

Nvidia press release · Tom’s Hardware · The Rundown AI · Techpresso · Axios AI+

  • Monday: the board opened an independent review after reports that Huang had been targeted by a foreign intelligence effort. It said it had “no reason to believe” that national security or Nvidia secrets were compromised (Fortune).

  • Thursday: Nvidia said Huang “has resigned as President, Chief Executive Officer, and a member of the Board, effective immediately” (Nvidia press release).

  • The interim: Colette Kress, CFO since 2013, is interim CEO and keeps the CFO job. Harvey C. Jones is the board’s first-ever chair, and a search is under way (Nvidia press release, The Rundown AI).

  • The market: shares closed down about 4% on Thursday after a much deeper drop during the day (The Rundown AI, Generative AI).

  • The staff: an internal post drew more than 1,600 comments, many urging the board to bring Huang back, WIRED reported (Techpresso).

  • The search: the board has begun approaching outside candidates, with no offer made, the Financial Times reported (Techpresso).

  • The structure he left: about 60 direct reports and no clear second in command (The Rundown AI, Axios AI+).

Why it matters: The most valuable company in AI had no named successor for the person it depended on most. Most enterprise AI programs have the same gap, one level down.

🤝 PARTNER SLOT
This space is reserved for one enterprise AI partner each week: AI security, resilience and agent controls for the leaders who read this briefing. Sponsorship options and media kit →

🔌 One supplier: outages and the checkbook

Techpresso · Axios AI+ · Generative AI

  • Microsoft 365: an outage hit Outlook, Teams and Copilot from Thursday into Friday. Downdetector tracked complaints in 74 countries, with Outlook reports peaking near 138,000 (Techpresso).

  • The cause: Microsoft blamed an overwhelmed process that handles secure server connections, not a cyberattack. The outage also broke OpenAI’s and Anthropic’s Microsoft connectors (Techpresso).

  • The same Friday: the FAA halted US flights for more than four hours. Officials believe a software update to a private network was the likely cause, and said neither the Microsoft outage nor the FAA’s AI systems played any part (Techpresso, citing CBS News).

  • Nvidia as a bank: Nvidia “has made $530 billion of off-balance-sheet commitments and guarantees,” according to SemiAnalysis, and the industry “remains highly dependent on its checkbook” (Axios AI+).

  • Switching got faster: Anthropic says its Mythos model helped run certain workloads on AMD chips in about three days, compared with six to 12 months in the past (Axios AI+).

Why it matters: Your AI assistants inherit the outages of everything they connect to. List what stops when one provider goes down, and what part of your plan a vendor is financing.

⚙️ One default: the setting nobody changed

The Decoder · Generative AI · The Deep View · Techpresso

  • Claude Code: security firm Apiiro found 2,539 secrets across 1,994 public projects, saved to a settings file when developers clicked “Yes, and don’t ask again” (Generative AI).

  • Grok Build: Claroc found 35,578 of 37,571 apps open to anyone without a login. On its test account, the publish dialog preselected “Anyone with the link” (The Decoder).

  • Agents that buy anyway: in Visa’s benchmark, 22 AI models passed 57.2% of shopping runs, and 54% of the failed runs still ended in a purchase (The Deep View).

  • Code nobody read: Amazon released AWS Emulator, written entirely by AI agents. “Our team hasn’t written or reviewed any of the code,” wrote AWS engineer Matthew Goldberg (Generative AI, single source).

  • The merchant’s answer: United Airlines’ revised Contract of Carriage lets it cancel bookings made by unauthorized AI agents (Techpresso).

Why it matters: None of these was a clever attack. Each was a default that favored speed. Review the defaults in every AI tool your teams use: what it saves, what it shares, and what it does when a check fails.

🧠 New models this week

The Rundown AI · Techpresso · The Batch · Generative AI

  • Claude Fable 5.2 and Mythos 5.2 (Anthropic): 20% cheaper and up to 30% faster than the 5.1 generation, at $8 in and $40 out per million tokens. Fable was held back in skills that speed up AI development: “We called for pacing, not pausing.” (The Rundown AI)

  • Gemini 4 Carbon (Google): its strongest model opened to developers and AI Ultra subscribers (The Rundown AI, Generative AI).

  • Gemini 4 Hydrogen (Google): $1.50 in and $7.50 out per million tokens, with adjustable reasoning levels, and 57% on Terminal-Bench 4.1 at high reasoning against 18.2% for Gemini 3.8 Flash. It is in Google Search for everyone (Techpresso, Generative AI).

  • Gemini Decision 1 (Google): answers only yes-no, multiple choice and rating questions, at $0.10 per million input tokens (Techpresso).

  • Muse Ember (Meta): its smallest model ties Claude Haiku 5.5 on Artificial Analysis’ index; a test task cost $0.14 against $0.41 (The Rundown AI).

  • GPT-6 Cyber (OpenAI): addressed 948 Linux kernel flaws that give full control of a machine, at around $1,280 each, down from around $3,100 in July. Restricted to approved customers (The Batch).

  • Qwen4-Max (Alibaba): a two-week preview with a one-million-token context (Generative AI).

Why it matters: Three price points opened this week: frontier, fast, and decision-only. The saving is in routing each task to the cheapest model that passes your own tests.

🧰 New products this week

Techpresso · The Rundown AI · Generative AI · The Next Web

  • ChatGPT Wallet (OpenAI): agents can buy things after an approval on each purchase. Built with American Express, Mastercard and Visa; the agent passes a per-transaction “payment token” and never sees the real card (Techpresso).

  • Claude Sheets, Whiteboards and Meetings (Anthropic): working spreadsheets, editable boards, and Claude taking notes in Google Meet calls, in beta for Team and Enterprise plans (Generative AI, The Rundown AI).

  • Businesses in ChatGPT (OpenAI): companies can claim a verified profile and add their own brand agent (Techpresso, Generative AI).

  • Control Room (OpenAI): a system that can cut off a research model’s output; cached permissions can delay a shutdown by up to 35 seconds (Techpresso).

  • Gemini glasses (Google): pre-orders from $395, in stores December 10. Faces are blurred before images go to Google’s servers for Gemini (Generative AI, Techpresso).

  • Tesla Cybercab: on sale at $29,990 to a few Austin-area customers; it won’t move without a $299-a-month Autonomy Subscription (Techpresso).

Why it matters: Anthropic is bringing work tools into the assistant; OpenAI is bringing businesses and payments into it. Decide which of those you want your staff and your customers inside.

💸 The money

Techpresso · The Batch

  • OpenAI: $10 billion in revenue in the September quarter, its first at that mark, with a $10.9 billion operating loss, Bloomberg reported (Techpresso).

  • Anthropic’s IPO is six times oversubscribed, according to a headline carried by Techpresso; the listing is expected in the coming week (The Rundown AI).

  • Search: Alphabet’s revenue rose 25% in the September quarter, but search revenue growth slowed to 12%, from 19% and 17% in the two quarters before (The Batch).

  • The price of a bug: hacking contests have paid $20,000 and $70,000 per Linux kernel exploit; GPT-6 Cyber wrote them “for a small fraction of that cost” (The Batch).

The Number: $530 billion. Nvidia’s commitments and guarantees, per SemiAnalysis. It is the week’s best measure of how much rested on one company.

Why it matters: The frontier still loses money, and its biggest financier just changed leaders. Budget for prices and terms that may move in 2027.


🎯 This week’s corners

🎓 Cert Corner: Your customer-support assistant depends on one model provider and one email connector. Which design best improves its resilience?

A) A larger server for the assistant

B) A second model behind a routing layer, queued retries, and a tested degraded mode

C) A longer timeout

D) Weekly backups of the chat history

🚩 Red Flag or Green Light? A team’s pipeline scans every push for secrets, hidden folders included. Agent settings files are excluded from git, and agents get short-lived credentials from a secrets manager. 🚩 or ✅?

🏛️ The Boardroom Call (governance): One executive holds your AI vendor relationships, the roadmap and the budget. There is no named deputy. What does the board ask for?

A) Nothing; the program is working

B) A named and prepared deputy, written records of vendor contacts and decisions, and a yearly succession review

C) An outside consultant to shadow the executive

D) Split the AI program across departments

Answers below. Tell us yours in the comments.

Strategic C-Suite Takeaways

  1. Name a deputy for the person who runs your AI program, and write down what only they know.

  2. Map what stops when one cloud, model or connector goes down, and test the fallback.

  3. Review the defaults in every AI tool: what it saves, shares and approves.

  4. Route by task: frontier, fast and decision models now have very different prices.

  5. Give agents their own spending identity, limits and approval, as the card networks just did.

The Week Ahead

  • Nov 17: Nvidia reports quarterly results, its first without Huang (Nvidia press release).

  • Nov 18: the date three House committee chairs requested for a closed-door briefing on the Nvidia matter (The Next Web).

  • This week: Anthropic’s listing is expected (The Rundown AI).

  • Dec 10: Google’s Gemini glasses reach stores (Generative AI).

What Else Happened in AI This Week?

  • OpenAI is breaking up its robotics team for the second time, moving researchers into its main research group. Techpresso

  • Google DeepMind’s agents mathematically proved the code behind Signal’s encryption in four weeks. Axios AI+

  • Google DeepMind released AlphaGenome 2, which predicts how DNA changes act inside specific cell types. The Rundown AI

  • Four nonprofits asked a San Francisco court to stop five AI companies from developing or deploying new models. Generative AI

  • Researcher Will DePue says patterns in published model weights point to restricted Nvidia chips at four Chinese labs; the data cannot prove smuggling. Techpresso

  • SpaceX’s Starship reached orbit on its 15th test flight and deployed 26 Starlink satellites, though the booster was lost. Techpresso


✅ Our take on this week’s corners

🎓 Cert Corner: B. Resilience comes from a second path and a plan for when both fail: route to another model, queue and retry, and tell the customer what still works. A bigger server (A) and a longer timeout (C) don’t help when the provider is down. Practice thousands more in DjamgaTech.

✅ Green Light. That is the fix for this week’s 2,539 leaked secrets: keep credentials out of commands, keep settings files out of git, and scan every push. 📘 Securing AI systems is a full domain of the AWS AI Practitioner exam: our 2026 guide.

🏛️ B. Nvidia had to build its succession in four days. A named deputy and written records cost little and can be done this quarter. A consultant (C) adds a second dependency, and splitting the program (D) removes the accountability you want to protect. Your org does it differently? Tell us. 📘 AI governance and strategy are at the heart of Google’s Gen AI Leader exam: our 2026 guide.

🚩 Red Flag or Green Light? is open for one sponsor. AI security companies: djamgamind.com/partners


🤝 Advertise with AI Unraveled & DjamgaMind

Put your enterprise AI brand in front of the CTOs, CISOs and engineering leaders who make AI buying decisions, daily, in English and French, on podcast apps, YouTube and Substack. Start by sponsoring one daily corner for four weeks.
Sponsorship options & media kit → · Email: etienne_noumen@djamgamind.com

🔗 Tools from the show

🎧 Prefer it commercial-free? Subscribe on Apple Podcasts.

⚗️ PRODUCTION NOTE: We Practice What We Preach.
AI Unraveled is researched and produced by Etienne Noumen, P.Eng. (Calgary, Canada) using a hybrid “Human-in-the-Loop” workflow.



Read original on Substack

Related articles

Level up your Tech Career with DjamgaTech PRO

Download the DjamgaTech PRO App on the Apple App Store to access hundreds of interactive prep questions, flashcards, and quizzes cleanly without ads.

Djamgatech PRO App
Download DjamgaTech PRO on the App Store 🚀

← All articles