01Claude does sheets and boards
Anthropic launched Claude Whiteboards (a shared board Claude can lay out from a spoken plan) and Claude Sheets (spreadsheets with tabs, formulas and drop-downs, editable together) in beta, and made Claude Dashboards generally available as Claude Data, adding certified queries and live sharing in beta. All three are on every paid plan; Sheets export to .xlsx or CSV, and boards and sheets can move on to partners like Miro, Lucid, Airtable, Asana, monday.com and Smartsheet.
Gemini 4 Carbon is out
Google made Gemini 4 Carbon, its strongest model, available to paying developers and enterprises through the API and to Google AI Ultra subscribers, two weeks after giving it to cybersecurity defenders. The introductory API price is $2 per million input tokens and $10 per million output tokens (rising to $4 and $20 after November 30), versus $8 and $40 for Claude Fable 5.2. Google says it ranks #1 on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench, and it is previewing a /swarm agent-team command in Antigravity.
Anthropic's IPO roadshow starts
Anthropic launched the roadshow for its IPO: 825 million Class A shares at an expected $70β78 each, listing on the Nasdaq under the symbol ANT. CNBC calculated a fully diluted valuation of about $1.4 trillion at the top of the range, and reported, citing people familiar with the plans, pricing on November 18, trading on November 19, about 20% of shares for retail investors, and a $26 billion concurrent private placement from seven companies including Nvidia, Microsoft, Google and TSMC.
38K findings, 217 fixed
Seven months into Project Glasswing, VulnCheck's Patrick Garrity tracked 442 CVEs attributable to Anthropic, about 0.5% of this year's CVEs, with two known to be exploited. Anthropic's own ledger listed 37,966 findings as of November 2, with 7,349 made public and 217 marked fixed; since September it added about 11,800 findings and 15 fixes. Of 34 Anthropic reports curl has published so far, 19 were judged not a vulnerability and 6 confirmed as low severity. VulnCheck says AI is a valuable tool but the triage burden is shifting to maintainers.
Notion's AI vendor got hacked
Notion says a security incident at its vendor Fireworks AI affected fewer than 1% of Notion workspaces: the exposed data is the text of Notion AI chat transcripts that Notion sent to Fireworks for its Insights analytics feature, without the workspace or user IDs. Fireworks had given Notion a contractual zero-data-retention commitment, but Notion says data was retained from the endpoints it used. Notion disconnected from Fireworks on November 3, disabled Insights, and recommends rotating any passwords or API keys pasted into Notion AI. Bloomberg reported Fireworks was breached twice in a few weeks, with Notion and Cursor among affected customers.
Apple's leaked local AI app
Leaker Aaron Perris posted screenshots and a description of Harbor, an unreleased app in private preview under Apple's Claris brand that turns a Mac into a private AI inference server for open-weight models, with in-app model downloads and the ability to pool unified memory across Thunderbolt 5-connected Macs to run larger models. Apple has not announced it (via 9to5Mac).
A voice AI that waits
Fish Audio released Live 3, a full-duplex voice model that keeps listening while it speaks: it can acknowledge with a short "mm-hmm" while you read out a number, keep going when you say "yeah," and stop when you actually cut in, deciding to respond in as little as 186 ms. Fish says the timing was learned from real human conversations rather than rules. It speaks English and is available on the Fish Agent Platform and API.
Big Walk's tiny AI slip
Nineteen jurors for the 2027 Independent Games Festival said they won't vote for games made with generative AI, including code. House House then told Aftermath that Big Walk's team occasionally used a chatbot to troubleshoot technical problems late in development, sometimes adapting suggested snippets, estimating it at well under 0.1% of the code, and apologized; it said Steam doesn't require disclosing this kind of use.
Disney picks Adobe for AI
At Adobe MAX, Adobe and The Walt Disney Company announced a multi-year alliance making Adobe Disney's official AI innovation partner, using Firefly Foundry to build custom generative models on Disney's characters, worlds and styles; Walt Disney Imagineering is already turning hand-drawn sketches into colored concept art. Disney dropped a planned $1 billion investment in OpenAI after OpenAI shut down the Sora app in March.
AI trader: 0 for 8
A Reddit user built a trading agent that has to pay for its own inference from a roughly $60 wallet, with the persona of a 70-year-old homeless robot on a park bench. After two weeks: 8 trades, 0 wins, $0.72 lost, $0 in tips, and one trade logged as "take-profit at +1198901.9%" that actually lost 7 cents. A separate signer process holds the key and enforces $3-per-trade and $10-per-day limits; at the current burn it has about 300 days left.
AI Found a Lost Babylonian Poem in Ancient Homework
GPT-6.1 Terra Just Dropped. I Actually Tested It.
GPT-6.1 Terra vs Claude Sonnet 5.5 vs Haiku 5.5: The Fastest AI Died First
AI Jailbreaking Explained: Does the Grandma Trick Still Work?
I Gave Opus 5.5, Haiku and DeepSeek the Same Game Prompt. They Made Triplets.
Opus 5.5 Motion Graphics: Zero After Effects, All Code
Can AI Crack Napoleon's Cipher? Claude Opus 5.5 vs GPT-6 Astra
I Locked an AI in 1900. Can It Beat Einstein to Relativity?
OpenAI and Claude Just Disproved a Famous Math Conjecture. Is P = NP?
