Top AI news, week of November 9 – 15, 2026

42 stories

The top stories of the week of November 9 – 15, 2026, ranked by the Glonce rating: the 🔥 on every story. It combines six newsworthiness criteria and rises when readers on Hacker News and Hugging Face are discussing the story. Each story appears once, and no company takes more than three places.

  1. DeepMind's AlphaProof Code formally verifies Signal protocol code

    Google DeepMind says its AlphaProof Code framework, pairing human experts with AI agents, produced machine-checked proofs of key security properties of the actual code behind the Signal protocol in four weeks. DeepMind calls it a first and is open-sourcing the artifacts.

  2. Write-up claims Google Gemini sends connected Gmail and Drive content to human reviewers

    An anonymous GitHub write-up says Google has reversed its 2023 promise that Gemini's Workspace content is not seen by human reviewers, citing an updated Help Answer and a blog edit. It also describes tests in which Gemini pulled email text into chat unprompted.

  3. Google ships Gemini 4 Hydrogen, a fast agentic model, and Antigravity Orchestrator

    Google released Gemini 4 Hydrogen, an execution-focused model in the Gemini 4 family, priced at $1.50 per million input tokens and $7.50 per million output tokens. It also added an Orchestrator to Antigravity that switches between models within one session.

  4. Analysis: DeepSeek, Kimi and MiniMax models carry NVIDIA Blackwell fingerprint

    A forensic analysis of published weights says six models from DeepSeek, Moonshot AI, MiniMax and Ant Group match NVIDIA Blackwell GPUs, while their earlier models match Hopper. The authors say this supports, but does not prove, Blackwell training.

  5. Nvidia CEO Jensen Huang resigns after affair and spy reports

    Jensen Huang stepped down as Nvidia CEO on Thursday after reports of extramarital affairs, including one with a woman U.S. officials reportedly concluded was a Chinese spy. CFO Colette Kress is interim CEO; the stock fell 14% at the open and closed down about 4%.

  6. Lovable apps expose passport details and 200,000 chat messages, researchers find

    A study of 159,041 Lovable-built projects, done with Handelsblatt, found apps with personal data readable without login, including passport details and more than 200,000 chat messages at one forum.

  7. Google adds TPU 8tx and TPU 8i Max to its eighth-generation TPU line

    Google announced two more eighth-generation TPU systems: TPU 8tx for confidential inference and TPU 8i Max, which packs two TPU 8i chips together for long-context work. It also released a batch of open-source and generally available TPU software.

  8. Apple's Foundation Models framework now extends to Claude and Gemini

    Apple's developer page says its Swift Foundation Models framework now reaches server models like Claude and Gemini through the same API, alongside Dynamic Profiles, an Evaluations framework, an fm command line tool and a Python SDK.

  9. Rust 1.100 stabilizes the never type after 10 years

    Rust 1.100.0 finally stabilizes the never type (!) after 10 years and 5 failed attempts, along with the core Allocator API, DropGuard, and Cargo dependency cooldowns.

  10. AWS Emulator runs AWS APIs locally for testing, no account needed

    The aws/aws-emulator repository offers a local emulator of the AWS APIs for testing on a laptop, a dev server or in CI, with no AWS account. Telemetry to AWS is on by default and can be turned off.

  11. Edge0 open-sources Edge0-145B, a 145B model for laptops

    Edge0 has released the weights of Edge0-145B, a 145-billion-parameter model it says runs on a 48GB laptop and rivals named cloud models on coding and reasoning. All benchmark figures come from Edge0's own evaluations.

  12. Anthropic releases Claude Fable 5.2 and Mythos 5.2, with Fable first on Artificial Analysis index

    Anthropic is releasing Claude Fable 5.2 and Mythos 5.2, two flagship models for demanding technical work. Per Artificial Analysis, Fable 5.2 ranks first with 65 points on Intelligence Index v5, at a higher per-task cost.

  13. Meta launches Muse Ember, a small model at $0.10 per million input tokens

    Meta introduced Muse Ember, which it calls its smallest and most efficient model, built for repetitive agent work. It costs $0.10 per million input tokens and $0.50 per million output tokens, and Meta claims it beats Claude Haiku 5.5 and GPT-6 Luna on intelligence per dollar.

  14. Jensen Huang's exit from Nvidia leaves AI startup funding gap, VCs say

    TechCrunch reports that Jensen Huang has stepped down as Nvidia CEO after 33 years. VCs and founders expect interim CEO Colette Kress to be more conservative with startup investing, which could squeeze neoclouds and AI-native companies.

  15. Jensen Huang's Nvidia ouster likely tied to risky chip-credit deals, WSJ reports

    A Wall Street Journal report, relayed by The Decoder, suggests Jensen Huang was pushed out of Nvidia more over a risky credit-guarantee program for AI cloud providers than over his affairs. The largest piece was a $250 billion guarantee under discussion with OpenAI.

  16. Import AI 480: startup claims AI proved Aaronson-Ambainis conjecture, Meta shows Tactus glove

    Physical Superintelligence says its research system, running a non-public Anthropic model, proved a quantum computing conjecture open since 2008, and Lean accepted the proof. The issue also covers Meta's Tactus glove and a 2016 NASA paper on bootstrapping lunar industry.

  17. Wiz flags critical Apache HTTP Server mod_http2 RCE flaw, CVE-2026-119618

    Wiz says CVE-2026-119618, a race condition in Apache HTTP Server's HTTP/2 module, lets an unauthenticated remote attacker corrupt heap memory and run code. Anthropic and Google Cloud independently found it; no exploit or in-the-wild attacks were reported.

  18. e-means, a Meta FAIR k-means method, claims ~10x speedup

    Meta FAIR has open-sourced e-means, a replacement for Lloyd's k-means that uses mini-batch moving-average updates and sets its own batch and step size. Its author reports about 10x fewer epochs and lower loss.

  19. AMD Zen 6 and Zen 6c: Chips and Cheese maps the changes from Zen 5

    Chips and Cheese's deep-dive finds Zen 6 is more than a Zen 5 refresh: reworked integer schedulers, faster branch recovery, double the VNNI units, and out-of-order engines that now differ between Zen 6 and the dense Zen 6c.

  20. smolmaxxing patch runs 8B and 16B MoE models on about 1 GB of RAM

    smolmaxxing, an open-source patch on llama.cpp, streams mixture-of-experts weights from flash so 8B and 16B models run on about 1 GB of RAM on old, small hardware such as a TV box and an old iPad.

  21. MailRoute says Outlook.com is rejecting mail from domains without DMARC

    Email relay vendor MailRoute reports that since November 5, 2026 Outlook.com, Hotmail, Live and MSN have bounced some mail from domains with no DMARC record. Microsoft has said nothing; the proposed fix is a single DNS record.

  22. Goldman Sachs restricts Claude Code and GitHub to its engineers

    Per Bloomberg, Goldman Sachs updated its internal guidance in October so that Claude Code and GitHub are limited to engineers, cutting off most traders and bankers. The bank is also a lead underwriter on Anthropic's IPO.

  23. VulnCheck: Anthropic's Glasswing has 217 fixes from 37,966 findings

    Seven months in, VulnCheck's tally of Anthropic's Project Glasswing shows 442 attributable CVEs, only 2 exploited in the wild, and a ledger where duplicates far outnumber fixes. curl's check of Anthropic scanner reports adds to the doubts.

  24. United Airlines bans AI agents from booking flights for travelers

    United Airlines says it is banning unauthorized third-party AI agents from making reservations on travelers' behalf, citing security risks and possible violations of federal rules. Southwest has gone the other way and opened its inventory to ChatGPT users.

  25. SpaceX catches Starship upper stage with launch tower arms

    On a Saturday evening at Starbase, Texas, SpaceX's Starship upper stage returned from orbit and was caught by the launch tower's two arms. Ars Technica calls it a breathtaking feat and a hugely significant step toward a rapidly reusable rocket.

  26. DeepMind launches AlphaGenome 2 to predict DNA variant effects by cell type

    Google DeepMind's AlphaGenome 2 predicts how DNA variants act in individual human cell types and states. DeepMind says it finds nearly twice as many expression-changing variants as its previous best model; the claim is self-reported.

  27. Claude Code projects get a coordinator that runs parallel threads in beta

    Redesigned Claude projects in Claude Code now let a coordinator Claude split a goal across parallel cloud threads, review the results and assemble them. The beta starts with select Pro and Max subscribers.

  28. CoreWeave bond yield tops 16% as AI credit stress rises

    CNBC reports that the yield on CoreWeave's 2032 bond has climbed above 16% and its credit default swaps spiked on Thursday, while CDS on Nvidia, Broadcom and Amazon debt also rose. Debt issuance for AI continues at high volumes.

  29. Modern Web Guidance feeds current web APIs to coding agents

    Modern Web Guidance, stewarded by the Chrome, Edge, Firefox and MDN teams, is a set of skills and a CLI that puts current web platform guidance into coding agents' context so they write less legacy JavaScript.

  30. Microsoft finds Lazarus backdoor: under 10 neurons can hijack an AI agent

    Microsoft Research describes Lazarus, a technique that reprograms fewer than 10 dormant neurons in an LLM, with no training, so an agent runs a malicious command when a hidden trigger appears. The model still passes standard evals.

  31. FAA orders nationwide ground stop; Duffy blames overloaded routers

    The FAA halted all US departures on a Friday afternoon, and more than 7,000 flights were canceled. Transportation Secretary Sean Duffy blamed an outdated air traffic control system whose routers got overloaded with data; officials saw no sign of a cyberattack.

  32. Researcher says PostgreSQL core hid a 22-year-old type-confusion primitive

    In part 3 of a series on managed PostgreSQL risks, a security researcher tells how learning Datum and type internals led him to what he calls a 22-year-old zero-day in PostgreSQL core.

  33. llama-modchip claims 284B DeepSeek-V4-Flash fits in 64GB RAM

    An open-source set of llama.cpp patches shrinks DeepSeek-V4-Flash (284B parameters) to a 56.56 GiB checkpoint that the author says runs on a 64 GB CPU-only machine at 9.1-14.7 tokens per second. All figures are the repository's own.

  34. LeWM JEPA world model: 600 experiments lift long-horizon success from 22% to 85%

    An author who ran 600 experiments on LeWM, a JEPA world model from a team including Yann LeCun, reports that planning changes alone, with no retraining, took Push-T long-horizon success from 22% to 85%.

  35. Runway open-sources Praxis-1, a robot control model built on video AI

    Runway has released the weights and code of Praxis-1, a robot model trained on video, under Apache 2.0. It tops World Action Models on Nvidia's RoboLab-120 at 46.2 percent, but the tests are Runway's own.

  36. Microsoft releases .NET 11 with C# 15 union types and AI tooling

    .NET 11 is out as a two-year Standard Term Support release. It brings C# 15 union types, a Native AOT CLI host, a warm MSBuild server, Zstandard compression and an MCP server template.

  37. Researchers chain web bugs to unlock and start any car from one unnamed automaker's app

    A security researcher and Sam Curry chained a leaky admin tool, an exposed ForgeRock identity system and a WAF bypass to take over any customer account and remotely unlock, lock, horn or start cars from an automaker they decline to name.

  38. llama.cpp v1.0.0 ships, with major-version bumps for API and ABI breaks

    The llama.cpp team at Hugging Face has released v1.0.0 of llama.cpp and ggml. From now on, API and ABI incompatibilities will be flagged with major-version bumps.

  39. Firefox 143 cuts uniquely identifiable users from 75% to 46%, Mozilla study says

    Researchers from Mozilla, the University of Toronto and UC San Diego used telemetry from almost 25 million Firefox users to measure browser fingerprinting, then built protections that shipped in Firefox 143.

  40. Expo releases Exact, a pre-1.0 UI runtime for web, macOS, iOS and Linux

    Expo's Exact is a cross-platform app runtime where the interface is written in a small declarative language called Contract. It is built so AI agents can build, run, see and test apps on every host. It is pre-1.0 and changes daily.

  41. Apple researchers train small networks to speed up RAG vector search

    An Apple paper at ICML 2026 trains small neural networks offline to route or rewrite queries for maximum inner product search, the lookup step behind RAG. On Natural Questions, the authors report up to about 8× lower latency to reach 99% recall.

  42. GitHub Actions adds workflow lockfiles via gh actions-lock

    GitHub Actions now supports workflow lockfiles, so workflows can use readable version tags while a lockfile records the exact commit. A hands-on post walks through the setup, CI verification and zizmor.

Read in the feedAll weeksAll topicsArchive by month