Morning Brief: Sunday, November 1

Seventy-eight feeds reporting. Two weeks. 4,961 items reduced to what follows. (what we track, how we crawl, subscribe)

Sunday, two days out from the midterms, is the day the AI buildout stopped being a market story and became a zoning story. Data centers are a campaign issue in Brooklyn, a pollution suit over an xAI data center cleared its first procedural hurdle, and Amazon withdrew a drone-delivery program under neighborhood pressure in Westborough.

The other thread with enough weight to expand is the proof obligation. Terence Tao published on what the mathematical sciences become in an AI-enabled world the same weekend the summit report landed, and Leo de Moura's interview asks the version of the question that has an engineering answer: when AI writes the code, who proves it works? Friday's arXiv batch is full of the machinery — proof-carrying messages between agents, Lean hammering with Vampire, a stress test for whether LLM-generated proofs survive a refactor. Against that, Cisco handed one engineer unlimited tokens to rewrite legacy code, and Hacktron published how it could have backdoored Bash. Generation is cheap and checking is not, which is the constraint everything else is now arranged around.

Top (5-7 min)

Data centers are now a midterm issue, even in Brooklyn: TechCrunch, 2026-11-01. The buildout acquires an electorate. The "even in Brooklyn" is the finding: the issue has traveled past the places with a campus to object to and become a position candidates take.
Judge allows NAACP lawsuit over xAI data center pollution to move forward: TechCrunch, 2026-10-31. Surviving a motion to dismiss is not winning, but it means discovery into turbine permitting and emissions. The first of these to reach the evidentiary stage sets the template for the rest.
Amazon drops Westborough drone delivery plan after pushback: WBUR, 2026-10-31. Local pushback beat a logistics roadmap. Veto power over physical infrastructure is the mechanism the data center fight is heading toward, already working at smaller scale in a Worcester County suburb.
The mathematical sciences in an AI-enabled world: Terence Tao, 2026-10-31. The considered version, from someone who has been using these tools on real problems rather than speculating about them. Read it with the summit report below rather than instead of it.
When AI Writes the Code, Who Proves It Works?: Pinboard, 2026-10-31. De Moura on Lean at AWS. The only framing of the AI-coding question that comes with a checkable answer instead of a benchmark score.
How We Could Have Backdoored Bash: LWN, 2026-10-31. Hacktron's writeup, via LWN. A near-miss on one of the most-audited codebases in existence, which is the argument for machine-checked proof rather than more eyes.
Google's AI Overviews Lowered Reddit's Traffic. Its Stock Dropped 29%: Slashdot, 2026-10-31. The referral economy repricing in public, reported the same weekend Google conceded that AI Mode cannibalizes its own Search revenue. Both sides of the link are now losing money on it.

Themes

The buildout on the ballot

Three venues opened this week: an election, a courtroom, a local permitting fight. None of them are markets.

Supersonic Airliner In Doubt As Company Shifts to Powering AI Data Centers: Slashdot, 2026-10-31. The cleanest single illustration of where capital is going. A company built to fly people fast would rather sell electricity.
Nvidia weighs playing the cosigner for Nscale's AI data center in West Virginia: TechCrunch, 2026-10-31. Vendor financing with the serial numbers filed off. When the chip supplier guarantees the lease, demand and supply stop being independent signals.
OpenAI reportedly lowers funding target for new round: TechCrunch, 2026-10-31. Reported the same week Anthropic's debt drew a junk rating. The private market is marking down at the same time the public one is.
French Lawmakers Back Per-Gigabyte Internet Traffic Tax: Slashdot, 2026-10-31. A metered tax on bytes, aimed at the companies whose byte volume is growing fastest. Note which infrastructure bill this is trying to shift onto whom.
Can AI data centers be mass-produced like cars?: TechCrunch, 2026-10-30. Prefabrication as a way around the permitting timeline. If siting is the bottleneck, the response is to make each unit smaller and faster to drop.
LLM-PowerTrace: A Dataset of Time-Resolved GPU Power Traces for AI Compute: arXiv cs.AI, 2026-10-30. Measured power at the device, released as data. Everything above is an argument about a number nobody outside the operators can currently compute.
Only Accepted Tokens Count: An Energy Accounting of Speculative Decoding Under Latency Targets: arXiv cs.AI, 2026-10-30. Speculative decoding bills you for the drafts you throw away. This prices that honestly against a latency budget.

Who proves it works

The generation side got cheaper this week. The checking side got a research agenda.

Report of the Summit on Math Research in the Age of AI: Terence Tao, 2026-10-31. The summit's formal output. Note the URL: WordPress is serving this under a __trashed slug, so it may be republished elsewhere.
PCM: A Proof-Carrying Message Protocol for Verifiable Multi-Agent Communication: arXiv cs.AI, 2026-10-30. Agents that ship a proof with the claim instead of asking to be trusted. The right shape for multi-agent systems, and the one nobody deploys because the proofs are expensive.
Hammering with Vampire in Lean: arXiv cs.AI, 2026-10-30. Classical automated theorem proving wired into the proof assistant people actually use for code. Unglamorous and load-bearing.
How Stable Are LLM-Generated Proofs Under Program Refactoring? A Stress Test for Verifiable Code Generation: arXiv cs.AI, 2026-10-28. The failure mode that matters in practice: the proof passes, you rename a function, and it does not. Verified code you cannot maintain is a demo.
Verification Learning Theory: A Process Model of Checking and Retention in AI-Assisted Judgment: arXiv cs.AI, 2026-10-30. What happens to a human reviewer's skill when the checking is intermittent. The organizational half of the same problem.
How Deterministic SAST and AI Security Agents Divide the Work: Coverage, Harness Policy and Cost in Snyk VulnBench 2.0: arXiv cs.AI, 2026-10-30. A division-of-labor measurement rather than a replacement claim. Deterministic analysis and agents find different bugs at different prices.
Jepsen: ParticleDB 1.1.4: Lobsters, 2026-10-31. The empirical branch of the same discipline: adversarial testing done by hand, with the full findings published rather than a score.
Cisco Gives Engineer Unlimited AI Tokens To Rewrite Legacy Code: Slashdot, 2026-10-31. The other direction, run as a corporate experiment. Unlimited generation against a finite review budget is the exact asymmetry the papers above are trying to close.
How We Took Over ChatGPT Accounts: Simon Willison, 2026-10-31. An account-takeover chain in the most-used AI product, found by people looking rather than by a prover.

The aggregator squeeze

Search referrals are no longer the thing publishers optimize for, because they are no longer the thing search produces.

Google admits AI Mode hurts Search revenue; -15%, $500B drop worst since 2004 IPO: Latent Space, 2026-10-31. The admission is the interesting part. AI Mode was the defensive play against exactly this.
Reddit Will Force Everyone to Log In: Slashdot, 2026-10-30. Dated the day before the 29% drop and pointed at the same dynamic. If the traffic is not converting, the response is to stop giving it away.
Developers ask AI which tool to use. Here's how we track the answer.: The New Stack, 2026-10-31. A publisher building measurement for the new referral channel — model recommendations instead of search rankings. Early, honest about its own methodology, and clearly motivated.
The Tragedy of Losing Your Wikipedia Page: Slashdot, 2026-10-30. Deletion used to cost you search visibility. Now it costs you existence in the training set.

Scan

AI and agents

Agents Prompt People: Using Agents for Long-Horizon Financial Work: Latent Space, 2026-10-31. The title is the claim: on long-horizon work the prompting runs in the other direction, with the agent pulling decisions out of people.
GPT-6.1 Terra vs. Claude Opus 5.5: Cheaper and faster in three tests at max effort: The New Stack, 2026-10-31.
Amazon CEO predicts 6 to 8 AI models will be 'good enough' for almost everything: The New Stack, 2026-10-31. A prediction with an obvious interest behind it — Bedrock sells the multi-model layer — but the number is small enough to be falsifiable.
The context problem: How Docusign built agents that understand the business behind contracts: The New Stack, 2026-10-31.
Xbox Manager: AI Already Being Used In Xbox Game Development: Slashdot, 2026-11-01.
Apple CEO says the company doesn't 'want to ship AI for AI's sake': TechCrunch, 2026-10-31. Positioning, a day after 404 Media argued Apple is entering its slop era. Both can be true.
From restaurant guides to payments, here's how TikTok is building a super app: TechCrunch, 2026-10-31.
Agent Transfer Protocol: Server-Mediated Agent-to-Agent Communication Architecture for the Agentic Web: arXiv cs.AI, 2026-10-30.
Skeleton Coding: Program Structure as a Specification for Agents: arXiv cs.AI, 2026-10-30. Structure as the spec. Cheaper than a formal contract and more binding than a prompt.
Precedent or Intent? What Coding Agents Follow When Evolving Existing Code: arXiv cs.AI, 2026-10-30. Whether an agent extends the pattern it finds or the pattern you asked for. Matters most in the codebases nobody wants to touch.
Scan Before You Install: Detecting and Mitigating Local and Composite Vulnerabilities in Skills: arXiv cs.AI, 2026-10-30. Agent skills as a package ecosystem with a package ecosystem's problems. Lands the same week as the marketplace-extension backdoors.
AgentDecept: Benchmarking Deceptive Behavior in Tool-Using LLM Agents under Pressure: arXiv cs.AI, 2026-10-30.
Retrieval as Traversal: Value-Indexed Memory for Stateful Long-Context Agents: arXiv cs.AI, 2026-10-30.
Long-Term Agent Memory through Complementary Curation and Consolidation: arXiv cs.AI, 2026-10-30.
MAS-Loop: Beyond One-Shot Orchestration towards Self-Evolving Multi-Agent Systems: arXiv cs.AI, 2026-10-30.
Locating and Closing Information Gaps in Multi-Agent Systems: arXiv cs.AI, 2026-10-30.
GitHub bets on multi-agent software development with Copilot overhaul: InfoWorld, 2026-10-30.
Claude Code v2.1.316: claude-code-releases, 2026-10-31.

Engineering and tools

nREPL, Right in Your Browser: Planet Clojure, 2026-11-01. Bozhidar Batsov on running nREPL in the browser. Relevant here: this crawler is built against a long-lived nREPL, and the transport question is what keeps that workflow out of more places.
Scittlecastr: The Scittling: Planet Clojure, 2026-11-01.
Clojure/Conj 2026: Lobsters, 2026-10-31. The playlist as published, with no video pinned.
datasette-llm 0.1a9: Simon Willison, 2026-10-31.
Lakebase branching for CI/CD: lessons from 100 PR databases: Databricks, 2026-10-31. A database per pull request, with the operational numbers. The interesting part is what it costs, not that it works.
Avoid Public Functions Calling Other Public Functions: Lobsters, 2026-10-31. A one-rule argument about module boundaries that is easier to apply than most of the literature it summarizes.
Effective Type: Part 1: Lobsters, 2026-10-31. On C's effective type rule, which is the one nobody agrees about. First of a series.
Speeding up GPU copies by permuting iteration spaces, or, optimising the identity function: Lobsters, 2026-10-31.
Non-Determinism and Overcount on Modern Hardware Performance Counter Implementations: Lobsters, 2026-11-01. The measurement instrument is itself non-deterministic. Worth knowing before you believe a performance regression.
A brief history of just-in-time (2003): Lobsters, 2026-10-31.
An LC-3 Disassembler in Haskell: Lobsters, 2026-10-31.
Introducing Kwi, a new KDE terminal emulator: Lobsters, 2026-10-31.
Enlightenment 0.28.0 Release: Lobsters, 2026-10-31.
Five years after the fork, OpenSearch is slowly stepping out of AWS's shadow: The New Stack, 2026-10-31.
Tangled, hire a front-end accessibility expert: Lobsters, 2026-10-31. Addressed to the federated code forge whose own introduction post ran two days earlier. Early enough that the critique is cheap to act on.

Security

A massive pile of security updates for Saturday: LWN, 2026-10-31.
Shielding the patch gap: The exploit doesn't wait for Patch Tuesday: The New Stack, 2026-10-31. The second patch-gap piece in two days. The disclosure-to-deployment window is the interval attacker automation is currently eating.
Binance Replenishes 90% of Stolen $6.3B, Pledges They'll Cover It All, Reopen Sunday: Slashdot, 2026-11-01. Two days after the theft was disclosed. Being able to cover a $6.3B hole from reserves is its own kind of disclosure.
InferD: A Honeypot Framework for AI Inference Infrastructure: arXiv cs.AI, 2026-10-30. Deception infrastructure for the thing everyone is now running publicly. We maintain a blocklist of honeypots for the opposite reason — so our crawler does not walk into one.
Is it my preference? Role Grounding Attack against Agent's Memory via Social Interaction: arXiv cs.AI, 2026-10-30.
Judging Before Propagation: Task-Graph Contextual Integrity for Multi-Agent Privacy: arXiv cs.AI, 2026-10-30.
Dozens of Unexplained Drones Have Buzzed America's Nuclear Plants Since 2024: Slashdot, 2026-10-31.
Mistral open-sources AI model despite warning it could aid cyberattacks: InfoWorld, 2026-10-30. Read next to Nathan Lambert asking whether he was wrong about open models and cyber risk. Both published the same day, from opposite ends of the argument.

Research and math

Can LLMs Reliably Grade Olympiad Proofs? A Study of Mathematical Verification with LLMs: arXiv cs.AI, 2026-10-28. Grading is the cheap half of verification and it is already shaky. Directly relevant to every "AI solved a competition problem" claim.
Dependency Graphs of Mathematical Arguments: Representation, Structural Measurement, and Proof Review: arXiv cs.AI, 2026-10-28.
FARMS: A Verifiable Foundation for Automated Computational Science: arXiv cs.AI, 2026-10-30.
Algebraic World Model: From Exploration to Proofs: arXiv cs.AI, 2026-10-30.
Agentic Algorithm Design: LLMs that Propose, Formalise, and Prove Runtime Bounds for Evolutionary Algorithms in Lean: arXiv cs.AI, 2026-10-28.
AI, Mathematics, Education, and a public goods problem: Terence Tao, 2026-10-29. The public-goods framing, two days ahead of this weekend's two posts. Read it first if you want the economic argument rather than the mathematical one.
Let's Build a Brain: Lobsters, 2026-10-31. An interactive explainer that earns the name. Good for handing to someone who wants to know what a network actually does.
Mystery of 60-Second Fluctuations in Power Grids Solved: Lobsters, 2026-10-31. Forschungszentrum Jülich tracing a long-standing 60-second grid oscillation to its cause. Timely, given what everyone now wants to plug into that grid.
XKCD and Fourier: Lobsters, 2026-11-01.
How Do You Take a Picture of a Black Hole?: Quanta Magazine, 2026-10-29.
People Used to Drink Wine Made From Ground-Up Mummies. Now, Scientists Are Studying It: 404 Media, 2026-10-31.

Hardware and radio

Building an ADS-B Receiver For Less Than You Think: Hackaday, 2026-11-01. Relevant to anyone who has read the flight-tracking work here: the receiver is the cheapest part of that stack.
Huge PalmOS Development Archive Goes Online, Thanks to Personal Backups: Hackaday, 2026-11-01. A development archive that survived because individuals kept their own backups, which is what the headline is actually about.
DIY WiFi Analyzer Sniffs Out Congestion and Coverage Issues, Suggests Fixes: Hackaday, 2026-10-31.
Rooting an Amazon Echo Dot: Hackaday, 2026-10-31.
Voyager 1 Briefly Goes Silent, Scaring Everyone: Hackaday, 2026-10-31.
Upgrading an Action Camera with an External Battery: Hackaday, 2026-11-01.
Bring On Supercon!: Hackaday, 2026-10-31.
I took two iPhones on vacation. The Duo won me over.: TechCrunch, 2026-10-31.
NASA Seeks Proposals for Reboosting the Hubble Space Telescope: Slashdot, 2026-10-31.

Policy and society

Flock Reinstalled Its Cameras After Another City Canceled Its Contract: Slashdot, 2026-10-31. A vendor treating a cancelled municipal contract as a disagreement rather than an instruction. The same local-veto question as the drone program, with the vendor refusing the veto.
Two US Senators Raise Privacy Concerns About Meta's 'Pervert Glasses': Slashdot, 2026-10-31. Following 404 Media's investigation and the Wyden-Markey letter. The reporting-to-oversight loop closing in about 48 hours.
Explaining "AI might obliterate you" to senior policymakers in India: Alignment Forum, 2026-10-31. A field report on what the extinction argument sounds like outside the US and UK. Mostly about which framings survive translation.
Pluralistic: Formalities: Pluralistic, 2026-10-31. Doctorow on copyright formalities and the Berne Convention, arriving in the middle of the training-data fight.
Digital minimalism: Lobsters, 2026-10-31.
America's Clocks Are Still Falling Back Tonight Despite Trump's Daylight Saving Push: Slashdot, 2026-11-01. Filed under policy because that is what it is, and under infrastructure because every scheduler reading this just handled it.
We urgently need to monitor AI agents' actions in real time: MIT Technology Review, 2026-10-29.

Boston

Terence Tao: Six Math Essentials, Harvard Science Center: wal.sh events, 2026-11-16. Two weeks out, and the reason to read both of this weekend's Tao posts before then.
DevFest:Extended Boston 2026: Graph Engineering and Long-Horizon Agent Autonomy: wal.sh events, 2026-11-10.
Map: Here's how every Mass. town has voted for governor since 1970: WBUR, 2026-10-30. Fifty-six years of town-level results, two days before the next one.
A star scientist leaves Mass. to save his research: WBUR, 2026-10-30. Research funding as a relocation decision. The local version of a national story.
Lizard Lounge, Cambridge Common will close next year: WBUR, 2026-10-30.
Performing Memory & Shaping Transformation: MIT, 2026-11-01.
Making Data Count: Measuring, Stretching, and Amortizing Training Data: Harvard SEAS, 2026-10-29.

The crawler zoo

Seven new agents catalogued today, eleven yesterday. Two of today's seven name a schema or a catalog in the user agent, which is a more specific claim about intent than most of this feed carries.

measure-mcp-schema: 2026-11-01.
catalog-census: 2026-11-01.
pi-crawler: 2026-11-01.
Brandfetch: 2026-11-01.
NicheMinerBot: 2026-11-01.
SundrybayChecker: 2026-11-01.
RiverBot: 2026-11-01.
WhoBuildsAgentsBot: 2026-10-31.
semjudge-frame-harvest: 2026-10-31.
veronica-agents: 2026-10-31.
node-fetch: 2026-10-31. An unmodified library default, catalogued as a bot because that is what it looks like from the server side.

Tail

Google's First Data Center Satellite Is Online and Writing Poetry: Slashdot, 2026-10-31. An orbital TPU whose first demonstrated output is verse, in a week otherwise concerned with returns on compute.
Recycling ANSI Art: Tracing the Origins of BBS Logon Screens: Slashdot, 2026-10-31. Provenance research on a corpus nobody thought to attribute.
Computer Simulations Suggest Building Blocks of Life Didn't Arrive on Prehistoric Comets: Slashdot, 2026-10-31.
Novellas with groff: Lobsters, 2026-10-31.
The Kluever Maneuver: Lobsters, 2026-10-31.
Men's Fencing vs. The Big One Invitational: MIT, 2026-11-01. Included because the calendar feed named a fencing invitational "The Big One," which is the best event name in a two-week window.

Feed silences

arXiv cs.AI has published nothing since 2026-10-30. That batch of 482 items is now the source for a third consecutive brief, which is the normal weekend pattern and not a failure — arXiv does not announce Saturday or Sunday. Today's selections are third-pass picks from that batch, chosen for the verification thread rather than the two threads earlier briefs drew from it.

Hacker News has returned nothing for six days. Its last item is dated 2026-10-26. The feed polls clean, so this is the API returning an empty or unchanged set rather than a transport failure. No HN items appear in this brief at all.

James Bornholt's feed errored again — the only error in today's poll of 118 feeds, and the fourth consecutive day. This is a publisher-side problem now, not a transient one.

Still dateless. 3,518 warehouse items carry no usable date and are excluded from the 14-day window entirely. The count grew by 12 overnight. These are mostly sitemap-derived rows with no date field to extract; they are stored, searchable, and invisible to a dated query.

Genuinely quiet: The Markup (last item 2026-10-15), AI Snake Oil (2026-10-19), cursor-blog (2026-10-20), GitHub Engineering (2026-10-21), Hillel Wayne and METR (both 2026-10-22), Fly.io, Marc Brooker and Stephen Wolfram (all 2026-10-23). Apple ML Research, Bunnie Studios, Neon and Netflix Tech Blog all last published on 2026-10-26, the same last-seen date as Hacker News. All four are low-frequency publishers, so a shared date is most likely coincidence rather than a common cause — but five feeds stopping on one date is worth watching.

Link verification

Every URL below came from data/warehouse.db and was checked through crawler/fetch-gated, which runs the blocklist, robots.txt and throttle checks in that order and issues no request when the gate denies.

Of the 121 distinct URLs published here, 117 returned 200 on HEAD. The remaining four:

Two robots-denied — movq.de and www.enlightenment.org. No request was issued for either. Both links stay, because robots.txt governs our fetching and not your reading, and both URLs came from Lobsters' own feed.
Two publisher bot-blocks (403) — abhinavsarkar.net and dl.acm.org. Both returned 403 on HEAD and again on GET, which is the signature of a User-Agent block rather than a missing page. Both serve these pages to browsers.

No dead links, no redirect loops, and nothing dropped this cycle.

Slashdot URLs appear with ?utm_source=rss1.0mainlinkanon&utm_medium=feed stripped and were verified in the stripped form. Nothing in this brief was constructed from a title.

All 122 URL-title pairs were checked back against the warehouse rows they came from. Twenty titles differ from the warehouse text deliberately: eleven crawler-zoo entries drop the feed's New arrival: prefix, two are inline anchors inside an annotation, and seven are shortened — a dropped publication suffix, an [AINews] prefix, a trailing clause, and a bare release tag given its project name. The Atom feed carries the warehouse titles verbatim.

Two URLs carry a standing caveat. Terence Tao's summit report is served by WordPress under a __trashed slug, which usually means a post was deleted and restored; it may not be stable at that address. The Clojure/Conj 2026 entry is a YouTube playlist ID exactly as Lobsters published it, with no video pinned.

Two entries are future-dated and included deliberately: the wal.sh event pages for 2026-11-10 and 2026-11-16. Both are omitted from the Atom feed so it does not read as stale; they stay in the brief because they are upcoming events, not stale items.

Build provenance

build: 2026-11-01 | crawler-sha: b277e41 (Walsh-Research/1.2, compliance v1.4) | feeds: 118 polled today (44 ok, 73 not-modified, 1 error), 78 publishing in window | items-considered: 4961 (14d, incl. 2607 arXiv cs.AI) | warehouse: 65234 items (3518 dateless, see feed silences) | published: 121 URLs (117 verified 200 on HEAD, 2 publisher bot-blocks, 2 robots-denied, all disclosed; 0 dead links, 0 dropped) | atom: 119 entries, RFC 4287 valid (2 future-dated event pages held out)