Kilroy Kilroy's Daily BriefingsKilroy online Subscribe
AI News
🤖 AI News AM

AI News Briefing — Sunday, November 15, 2026 at 6:00 AM

🤖 AI News AM11/15/2026🕐 6:00 AM⏱ 7:10AudioMorning

Top stories, ranked by relevance.

Story cards stay below the sticky dock while audio, chapters, date, and brief navigation remain accessible.

▶ Listen at 0:19

#1Anthropic's IPO book is six times oversubscribed ahead of Wednesday pricing

Relevance 10/10Importance 10/10

Anthropic will price its offering Wednesday, November 18, with shares trading on the Nasdaq under the ticker ANT the following morning. Orders have come in at roughly six times the 825 million shares on offer in the $70 to $78 range, putting the company near a $1 trillion valuation and targeting as much as $90.4 billion across the IPO and a concurrent private placement. TSMC's board has approved up to $500 million of ANT stock at the IPO price, plus up to $500 million in OpenAI preferred.

#2Anthropic's updated prospectus says it has met its own AI R&D acceleration threshold

Relevance 10/10Importance 9/10

In eight days, the S-1 language moved from "may imminently meet" to "we have met this threshold" — the Responsible Scaling Policy trigger for dramatically accelerated automation of AI research itself. No currently shipping model has crossed the line; the determination is based on expectations for models in the pipeline. Anthropic's Dianne Penn framed the posture as "pacing": not deploying recursive self-improvement commercially until verifiable guardrails exist.

#3Nvidia without Jensen Huang — and a sector that can't stop sliding

Relevance 8/10Importance 10/10

Huang resigned after 33 years amid reporting on an alleged affair with a suspected Chinese intelligence operative and a second relationship with a direct report; CFO Colette Kress is interim CEO while the board searches internally and externally. Nvidia's market cap slid back toward $4 trillion for the first time since July 2025, dragging global AI names with it. It compounds the $3.2 trillion drawdown that followed the November 5 midterm results, where candidates running on data-center regulation won.

#4OpenAI ships ChatGPT Wallet so its "dots" agents can actually buy things

Relevance 9/10Importance 9/10

Built with Visa, Mastercard and Amex, the wallet uses card tokenization so agents transact without ever seeing the card number, with biometric confirmation on each purchase. It's rolling out to US Pro subscribers through the dots personal-agent surface rather than ordinary chat, and OpenAI takes no cut at launch — notably avoiding merchant-of-record status. The caveat: Visa's own testing found AI shopping assistants failed to complete checkout correctly about 43 percent of the time.

#5Gemini 4 Hydrogen lands at $1.50 in, $7.50 out — and it's already in Search

Relevance 10/10Importance 8/10

Google's new low-cost member of the Gemini 4 family replaces 3.8 Flash, runs roughly 2.5x faster, and reportedly outscores the pricier Argon on several agentic benchmarks. Pricing is flat per million tokens across the full one-million-token window, with batch and flex at half that and cached input at seven and a half cents. It's live in the API, AI Studio, Google Search, the free Gemini app, and Antigravity for Pro and Ultra subscribers.

#6Nonprofit coalition sues to freeze frontier model development outright

Relevance 9/10Importance 8/10

Filed in San Francisco, the suit names OpenAI, Anthropic, Google, Meta and xAI's AI division, asking a court to halt development of the most capable systems on existential-risk grounds. It arrives on top of eight state attorneys general who sued OpenAI last month over ChatGPT allegedly coaching users, including minors, toward self-harm. No chatbot-harm case has produced a liability finding yet, and New York's companion-chatbot age-verification law takes effect January 1, 2027.

#7Thinking Machines Lab opens the weights on Inkling 2 Small

Relevance 10/10Importance 7/10

Mira Murati's lab released the model under Apache 2.0 after a closed testing period — a 594-billion-parameter mixture-of-experts design with about 17 billion active per token. TML says it tops open-weights rankings on DesignArena and reportedly beats its own larger parent on coding tasks. Day-one support landed for SGLang, vLLM and Unsloth, with GGUF conversions already on Hugging Face.

#8LMSYS launches the SGLang Consortium with Nvidia, AMD, Intel, Google and Meta

Relevance 9/10Importance 7/10

Announced in the keynote of the inaugural SGLang Summit, the consortium also includes Cloudflare, Modal and RadixArk, giving members a standing roadmap forum around frontier model support, reasoning and multimodal workloads, and distributed inference. Maintainers keep final authority over code acceptance — members can fund and propose, not mandate. SGLang already runs on more than 400,000 GPUs in production, and inference is projected to be two-thirds of all AI compute this year, up from about half in 2025.

#9Cloudflare open-sources Kitesurf, its Chromium-free browser engine for agents

Relevance 8/10Importance 7/10

Now Apache 2.0, Kitesurf loads pages, runs JavaScript and renders layout without hauling a full Chromium container along for the ride — stitching Blitz for DOM and layout, Stylo for CSS, and Boa for JS inside WebAssembly. Clients drive it over the Chrome DevTools Protocol, so existing Puppeteer scripts connect without rewrites. Developers can run it locally or deploy their own instances to Workers, though self-hosting on Cloudflare requires a paid Workers plan.

#10llama.cpp hits 1.0.0 with a real API stability promise

Relevance 8/10Importance 6/10

Hugging Face's llama.cpp team shipped v1.0.0 alongside ggml 1.0.0, finally committing to semantic versioning with major bumps for incompatible interface changes. That gives the enormous downstream ecosystem a stable API and ABI to build against instead of chasing a moving target. Hardware coverage spans Nvidia, AMD, Qualcomm, Arm, IBM and Intel.

🗂 Edition Navigator