Tap to play. It never plays by itself, and the captions are in the picture.
Google's newest frontier model goes to cyber defenders first. Google DeepMind announced Gemini 4 Carbon. Google says it scored 88.2 percent on a vulnerability test. Cyber defenders get it first. Microsoft revealed Azure revenue for the first time: 33.2 billion dollars, up 48 percent. On Surge AI's new tax benchmark, the best agent passed just 10.8 percent. Alphabet sank 10 percent after negative free cash flow of 9.07 billion dollars. LangChain says its ad agent cut cost per lead 30 percent, with humans approving campaign changes. Agents do the analysis. Code holds the safeguards. People sign off.
Made with AI from this Drop's sourced items. The facts come from the sources linked on each item. Script: Claude · Voice: OpenAI gpt-4o-mini-tts · Visuals: motion graphics
Google DeepMind says Gemini 4 Carbon scores 88.2% on SEC-bench Pro and resisted every evaluated prompt-injection attack. Access starts with a subset of cyber defenders; the API will cost $2/$10 per million tokens.
Why it matters: A new frontier option for security and long-running coding work, but builders can't use it yet. Ars notes Google gave no date for a wide release.
CNBC reports Microsoft's revenue grew 19% to $92.28 billion and capex was $45.2 billion, below estimates. TechCrunch says $10 billion of the quarter's revenue came from arrangements with OpenAI.
Why it matters: Microsoft's CFO says demand still exceeds supply. CNBC reports neocloud stocks CoreWeave and Nebius slid almost 5% after the lower capex figure.
Surge AI's DAYJOB: Tax tests agents on returns, workpapers and advice across eight areas of tax practice. Claude Sonnet 5.5 leads at 10.8% pass@1, ahead of Claude Fable 5.1 at 9.6% and Claude Opus 5.5 at 9.2%.
Why it matters: Surge says agents can miss issues clients never raised, like a looming sale deadline. A pass needs 95% of rubric criteria, and professional judgment remains out of reach.
CNBC reports Alphabet's free cash flow was negative $9.07 billion and revenue and earnings missed estimates. Cloud revenue more than doubled to $30.7 billion; Google says the Gemini app reached 1 billion users.
Why it matters: Investors are testing whether AI spending pays off. TechCrunch reports Google is taking AI Mode ads to international markets as Search growth slowed to 12%.
LangChain says cost per qualified lead fell 30% from June to August and paid media went from 0 to 20% of its marketing pipeline in six months. Its Slack agent routes proposed campaign changes through human approval.
Why it matters: LangChain moved calculations and safeguards into code, like a rule against cutting a top pipeline driver after one bad week; one report got about 40x cheaper.
Why it matters: Willison calls the numbers a stark representation of the impact of agentic engineering on GitHub, the platform where much of that agent-written code lands.
Simon Willison relays GitHub's Octoverse figures: commits more than tripled to 7.38 billion a month, mostly in the second half of the year, and new repositories doubled to 30.3 million a month.
Why it matters: Jellyfish told TechCrunch Anthropic-based tools make nearly a third of proposed code changes, with spend up 10x this year to about 2% of engineering cost.
TechCrunch reports managers can see a team's output, its cost and where workflows broke down, and set pooled group budgets. Anthropic says it won't use this data for model training by default.
Why it matters: A new way to reach users through AI assistants; TechCrunch says developers can track traffic, latency and error rates once their apps are live.
TechCrunch reports developers can connect apps through a cloud-hosted MCP server or Android AppFunctions, then register and submit them. Over 50 connected apps in Gemini will come to Search in the next few weeks.
Why it matters: The authors find LLM agents build far better than they play: TavernAgent answered 91% of play tasks, against at most 63% for frontier LLMs.
Researchers report AutoTavern's agents, with a human setting goals, built a game simulator and 23 versions of TavernAgent in 20 days. Its last 20 live games averaged a 1.60 placement with 80% first places.
Why it matters: Agents that act for people need to know when to ask. The authors find personalization lags well behind execution.
Researchers report iPrefBench, 130 underspecified tasks across 26 iOS apps, tests whether agents infer missing details from history or ask. Seven frontier agents struggled most to ask for what only the user knows.
Why it matters: Experts tell CNBC the duty to balance mission and profit could weigh on valuation and M&A, and makes damages claims against directors harder to win.
CNBC reports Anthropic aims for a valuation of $1.5 trillion or more as a public benefit corporation, and seven co-founders hold 50.1% of voting power through a single Class F share.
Why it matters: Oracle's AI buildout supplies capacity to OpenAI. CNBC notes it has issued $117 billion in long-term debt and is rated BBB-, the lowest investment grade.
CNBC reports Oracle shares jumped 11% after it projected 35% average annual growth and said it will turn free cash flow positive in the second half of fiscal 2028.
Why it matters: Samsung expects AI infrastructure investment to accelerate in 2027, with rising demand likely to further tighten memory supply.
CNBC reports operating profit of 107.44 trillion won, up 783.11% year on year. Samsung says it is the first global tech company to top 100 trillion won in quarterly operating profit.
Why it matters: Legal experts say revised terms may still not rule out AI training on news sites, The Verge reports, and senators from both parties have objected.
The Verge reports Meta's Andy Stone said it won't change policy without a full consultation with industry groups, which could run past November 20; existing terms stay in place until then.
Why it matters: The Art. 50(2) marking duty covers generative AI already on the EU market, so the date reaches existing products, not only new launches.
From December 2, per the EU AI Act: bans apply to AI making non-consensual sexual deepfakes of real people or child sexual abuse material, and generative AI on the market before Aug 2, 2026 must mark output.
Why it matters: The biggest gainer in our repo watch this run, aimed at teams pointing coding agents at Apple platforms.
Agent skills to research, clone, design, build, optimize and test native macOS apps with Swift and SwiftUI. Up 122 stars since our last check; MIT license; created October 24.
Why it matters: A security review aid for teams already running coding agents. It is a day old, so judge its maturity accordingly.
Point Claude Code, Codex or any agent at a repo to get an evidence-traced threat model; the repo lists 36 architectures and 29 methods. MIT license; created October 28.
Why it matters: Safety filters in a search layer can make answers less useful. The Hacker News thread drew 136 points and 65 comments.
Ars Technica found AI Mode pointed users to the FEC instead of naming the 2020 winner, while the Gemini app answered directly. A Google spokesperson cited a technical gap for 2020 and 2022 Senate races.
Why it matters: Chatbot builders serving young users in New York may face new state rules; we found no official bill text in this run's sources.
Posted by @GovKathyHochul on X: she says she signed new laws to protect children from dangerous AI chatbots. @NewYorkStateAG posted that the attorney general's office worked on the legislation.
Why it matters: Open-source projects are setting their own AI rules, so teams using coding agents should check a project's policy before contributing.
MicroZig's code of conduct, in effect from October 28, 2026, bans LLM-generated code and prose, LLM editing, translation and bug-finding, and talk of chatbot use. The HN thread drew 49 points and 61 comments.