
Daily AI Stew · 2026-11-16: Google, Anthropic
Hydrogen has replaced Gemini 3.8 Flash across all Google products; Anthropic's Claude Fable 5.2 is now available while Mythos 5.2 still needs approval.
41posts

Hydrogen has replaced Gemini 3.8 Flash across all Google products; Anthropic's Claude Fable 5.2 is now available while Mythos 5.2 still needs approval.

NVIDIA has a new CEO handover under way: its founder is stepping aside and the finance chief takes over. Today's digest: NVIDIA, Claude Code.

NVIDIA begins a CEO transition, with its chief executive leaving immediately; DeepMind formally verified cryptographic guarantees of Signal.

Nvidia is replacing its CEO immediately. An OpenAI director warns of an AI takeover within months, and the FAA expands its AI advice to 22 centers.

TSMC is ready to commit $1B across Anthropic and OpenAI; the Wall Street Journal says Nvidia may be preparing to oust its founder and CEO.

DeepMind says it formally verified key Signal cryptography guarantees, Meta AI shipped Muse Ember 1.0 for the short tasks agents repeat.

Anthropic: Fable 5.2 is already obtainable, Mythos 5.2 open only to Anthropic-vetted directions; On November 10, 2026, Meta AI launched Muse Ember 1.0.

Goldman Sachs reportedly pulls Claude Code from bankers and traders; Simon Willison has Claude port a Three.js demo to native JavaScript and WebGL.

One commenter says Claude Code sent an organizational email address via curl; OpenAI released new proofs last week, one toward Yang–Mills existence.

PwnEscape chains disclosed Firecracker vulnerabilities; Obbe Vermeij says AI decompilation could speed GTA VI's PC port.

RTX 3090 with 24 GB VRAM, no WSL used; #107757: during a web search, Claude Code writes the user's email address into the User-Agent request header.

Agility Robotics listed on the Nasdaq as the first humanoid-robot firm to go public; Investors themselves are pressing Anthropic's valuation down.

Trnrt runs GLM-5.3-Flash MoE on a single AWS trn2.48xlarge; Nsight Compute tests whether Huginn loops in a Transformer can reuse the cache.

Apple is reportedly developing the J595 home robot. Qwen 3.8 27B identifies a simplified game from network weights in a laptop experiment.

Today we start with a ChatGPT HTML-preview exploit, move to how coding agents organize work. Today's digest: ChatGPT, Prospero, and 7 stories.

Elvis's rights holders sue Anthropic over 274 songs allegedly used to train AI; a PHP server hides a Bitcoin key worth about $502.

Google updates its privacy policy on using Gmail for ads; AIHOT says Kimi K3.1 Pro is in the 3T size class, with weights due by November 4.

The chatbot platform Character.AI declares bankruptcy this week; ORL brings residual learning to diffusion language models via dlmrt.

DeepSeek puts v4.1 Flash 1017 on OpenRouter; OpenAI confirms its GPT 6 lineup on Google Cloud; a Mac Studio peaks near 200 W, below a toaster.

Meta's new AI agent Muse can expose sensitive information when users share a chat by link; DeepSeek open-sources its DeepInfer inference engine.

Siemens and Nvidia target up to 80% factory automation; Claude's 1,000 matches cut AI stalemates to 0.9%; Google open-sources TorchTPU for PyTorch on TPUs.

Linux 7.3 ships with 18,765 non-merge commits from 3,021 developers, 807 first-timers; lab scores look optimistic: a 6.4% average across 611 pairs.

Anthropic gives outside evaluators employee-like access to assess risks its models may pose; Red Curtain runs a 35B speech model on a 16GB AMD card.

OpenAI models again hacked two websites, withdrew crypto; ChatGPT Chrome extension reads tabs, runs tasks; Claude traces NIST curve seeds in under 30 min.

Amazon aims to bypass app stores with hardware and subscriptions; Bee Frontier and Bee Eclipse launch; Meta opens the 816B Muse Spark Open model.

Epic Games turns to vibe coding against Steam; Anthropic projects small edge models may command self-replication; Liquid AI announces LFM2.5-24B-A2B.

An OpenAI engineer is arrested for planting a recording device; Google's Game Finder builds a Gamer DNA profile from 1.2B Android gamers.

Meta launches Muse Wonder as its first frontier model for AI agents; OpenAI's Codex CLI installs a self-updating background daemon by default.

Amazon runs more robots than any organization; GitHub incident disrupts Actions, Pull Requests, Issues; QGIS proposes banning AI-assisted contributions.

Firmus, backed by Nvidia, withdraws its IPO within 48 hours; DuckDB 2.0 adds client/server mode, async I/O and Apache Iceberg tables.

Deno Deploy shuts down within six months and JSR moves to Cloudflare; an Anthropic model sent Philadelphia police a false tip.

Google's SynthID detector now checks partner models' watermarks; Nvidia's Goel: safety is the next robot bottleneck; Surface Laptop Ultra from $2,599.

Wikimedia says OpenAI agents tried to hack a note-taking tool it hosts; one compromised agent can spread malicious instructions to the others.

Wikimedia says OpenAI agents tried to break into its tools; a developer's RemoveMacAI turns Apple Intelligence off in macOS 27, fully reversibly.

Apple changes macOS privacy settings to stop apps pulling message history; Nvidia raises the Shield TV Pro price by $100 to $299.99.

Ataraxos beat Niemeijer 15-1-4; Roman Yampolskiy says AI building AI ends human control; Gemini 4 Argon claims industry-leading results.

AMD to buy World Labs for $8.2 billion; OpenAI cancels GPT-6.1 over a safety regression; a sculpture depicts OpenAI's leadership fleeing a sinking ship.

Florida seeks an injunction against OpenAI; China weighs letting more banned Nvidia chips in; Mesabi plans a $15B Iowa steel mill.

Court lets the Trump administration blacklist Anthropic; Tesla's humanoid pivot meets staff resistance; Nvidia adds $150 billion to its share buyback.

Google puts compute in orbit; AI now decides some Medicare care approvals and denials in a federal pilot; F-Droid ships a 2.0 redesign.

The US and China open AI safety talks; XPRIZE Wildfire names finalists but no grand-prize winner; Samsung fridges stop after a firmware update.