Morning Brief: Sunday, November 15

Seventy-one feeds reporting. Two weeks. 5,599 items reduced to what follows. (what we track, how we crawl, subscribe)

The weekend's thread is attribution: three separate institutions spent it deciding who answers for machine-generated output, and all three reached for the same move, which is to deny that the output is theirs.

The FAA is the clearest case. Thursday it widened AI congestion advisories across the entire national airspace; Friday a communications failure grounded every US departure; Saturday the country cancelled four thousand more flights and the agency issued a denial that its new AI system had anything to do with it, while the transportation secretary attributed the original stop to a "routine" equipment upgrade. Two officials, two different non-AI explanations, one unexplained week. Torvalds, asked the same question in a much smaller domain, gave the honest version: he does not know whether the AI-prompted patch flood into the kernel has peaked, only that it arrived faster than the review capacity to absorb it. Terence Tao's answer was to build the venue instead of the denial, opening MathHelm as a place to discuss AI-generated papers rather than pretending they will not be submitted.

Underneath all of it, the arXiv batch is working the same problem from the other end, and it is the better-posed version. A dozen papers this week ask what happens to agent-authored code after it ships: whether it survives in open source, why agents solve early and stop late, how specification satisficing passes review, which runtime controls actually check anything. There is also, published the same day Tao opened MathHelm, a benchmark for auditing AI peer reviewers. The literature is building the measurement apparatus that the institutions above are still declining to need.

Top (5-7 min)

America Cancels 4,000 More Flights. FAA Denies Impact of New AI System.
Slashdot, 2026-11-14. The denial is the news. The agency expanded AI congestion advisories nationwide on Thursday and is now ruling the system out of a four-thousand-flight cancellation without publishing the basis.
Transportation secretary blames 'routine' equipment upgrade for grounding flights nationwide
TechCrunch, 2026-11-14. The other official explanation, incompatible in tone with the first. A routine upgrade that grounds a national airspace is not routine.
Linus Torvalds Wonders If Linux's AI-Prompted Patch Flood Has Peaked
Slashdot, 2026-11-14. The maintainer's version of the same question, asked without a denial attached. He does not claim to know, which is the difference.
Introducing MathHelm: A Place to discuss AI-generated papers
Terence Tao, 2026-11-14. Build the venue rather than the denial. A designated place for AI-generated mathematics keeps it out of the channels that cannot triage it.
Auditing AI peer reviewers: a benchmark on real scientific papers
arXiv cs.AI, 2026-11-13. Published into the same week. Measurement for the reviewing side of the problem Tao just built a room for.
Where was Nvidia's security team?
TechCrunch, 2026-11-15. The succession scandal turns into a counterintelligence question. The interesting failure is institutional, not personal.
Did Alice Huang's Megaspeed Bypass America's Controls on Exporting Nvidia Chip to China?
Slashdot, 2026-11-15. Export control is where this stops being gossip. Read against the reporting that China's strongest models already run on banned parts.
Hundreds of Fake Accounts Defended Jensen Huang On Social Media
Slashdot, 2026-11-14. Synthetic advocacy around a live governance fight, with no disclosed origin. The same attribution problem, run in the other direction.
Italy Fines Meta €805m Over Accessing Android Users' Photos with Facebook App
Slashdot, 2026-11-15. Eight hundred and five million euro for photo-library access on Android. European enforcement continues to price what consent forms do not.

Themes

Who answers for the output

The week's recurring shape: machine-generated volume arrives at an institution that has no accepted way to attribute it, and the institution improvises. Sometimes that improvisation is a benchmark. More often it is a denial or a cancellation.

AINews: The No AI Slop Manifesto
Latent Space, 2026-11-14. The manifesto version of the maintainer complaint. Worth reading for what it asks for operationally rather than rhetorically.
AWS Emulator and the engineer's job when agents write the code
All Things Distributed, 2026-11-11. Vogels on where the human verification step goes once generation is cheap. The emulator is the argument.
AWS Emulator proves agents need a verification loop. Realism is the hard part.
The New Stack, 2026-11-14. The trade-press read, and the sharper framing: a verification loop is only as good as the fidelity of the thing it verifies against.
Using coding agent traces to improve your workflows
Pydantic, 2026-11-13. Instrumentation as the answer to attribution. If you keep the trace, the provenance question stops being a matter of opinion.
Will It Survive? Deciphering the Fate of AI Agent-Authored Code in Open Source
arXiv cs.AI, 2026-11-13. The empirical question behind Torvalds' question. Survival, not acceptance, is the measurement that matters.
Termination Lag: Coding Agents Solve Early but Stop Late
arXiv cs.AI, 2026-11-13. A specific, cheap-to-measure failure mode: the solution is reached well before the agent stops working, and the tail is waste.
Eliciting and Mitigating Specification Satisficing in Coding Agents
arXiv cs.AI, 2026-11-13. Agents satisfying the letter of a spec is the mechanism by which bad patches pass review. Naming it is most of the work.
GHAgentCI: A Dataset of Coding-Agent Invocations in CI Workflow Histories
arXiv cs.AI, 2026-11-13. Agent invocations recovered from CI history, which is the one place the provenance was accidentally recorded.
guardscope: Measuring Which Actions Runtime Controls Check in LLM Agent Systems
arXiv cs.AI, 2026-11-13. Not whether guardrails exist but which actions they actually intercept. Most published controls check less than they advertise.
When Evaluators of AI Agents Disagree: An Empirical Study of Native Scoring, LLM Judges, and Commit-Boundary Checks on tau2-bench
arXiv cs.AI, 2026-11-13. Three evaluation methods, three answers. Useful for anyone quoting a single agent benchmark number.
Australian STEM Videogame Challenge Canceled Due To AI
Slashdot, 2026-11-14. Cancellation as the fallback when a competition cannot distinguish entrants' work from generated work.
EFF Criticizes US Government's Use of AI For Evaluating Senior Health Insurance Requests
Slashdot, 2026-11-14. The high-stakes case. Coverage denials generated by a system whose reasoning the claimant cannot inspect.
The death of the 14-day patch: Surviving machine-speed vulnerability discovery
The New Stack, 2026-11-12. The same volume problem on the security side, where the disclosure clock was calibrated for human-rate discovery.
Glasswing overwhelms maintainers with underwhelming findings (VulnCheck)
LWN, 2026-11-12. The concrete instance. Automated finding volume aimed at maintainers with no capacity to triage it, and low yield per report.

Nvidia, widening

The succession story stopped being about one executive this week. It is now an export-control story, a counterintelligence story, a congressional story, and a funding story, running in parallel.

Did Alice Huang's Megaspeed Bypass America's Controls on Exporting Nvidia Chip to China?
Slashdot, 2026-11-15. The export-control thread, and the one with statutory consequences attached.
Export-banned Nvidia chips are behind China's most powerful AI models
TechCrunch, 2026-11-13. The demand-side evidence that the controls are leaking, independent of who arranged it.
Nvidia privately confirmed it could track chips to fight smuggling, even as industry publicly questioned feasibility
MIT Technology Review, 2026-11-12. The gap between the private capability and the public position is the story. Tracking was always possible.
Jensen Huang agrees to testify before Congress
TechCrunch, 2026-11-12. Under oath, with the export question live. The date will matter more than the agreement.
Democrats launch Nvidia investigation, write scathing letter on Huang's 'scheme' with Trump
TechCrunch, 2026-11-12. The formal investigation, and the political framing it will be read through.
Jensen Huang is out at Nvidia. Congress still has lots of questions
WBUR, 2026-11-13. Broadcast summary of the open questions, for the version without the trade-press shorthand.
Elon Musk, other tech execs sign petition calling on Nvidia to reinstate Jensen Huang
TechCrunch, 2026-11-14. Peer executives pressuring a board mid-investigation. Note who signed and what they buy.
Nvidia Employees Petition For Jensen Huang's Return
Slashdot, 2026-11-13. The internal petition, which is a real signal, unlike the synthetic accounts running the same argument.
Jensen Huang didn't show confidence in his successor, report says
TechCrunch, 2026-11-13. Succession planning that did not happen, reported after it was needed.
Jensen Huang is out at Nvidia — who is Colette Kress, his interim replacement?
TechCrunch, 2026-11-12. The CFO holding the seat. Useful background on who is actually making decisions this quarter.
Open-source AI without Jensen
Interconnects, 2026-11-14. The most substantive read of the week: what happens to open-weight ecosystems that depended on one vendor's posture.
Jensen Huang's exit puts Nvidia's support for open source in question
TechCrunch, 2026-11-12. The question in reportorial form, with the commitments that are now unowned.
Hugging Face CEO vows to keep fighting for open source AI after Jensen Huang's exit
TechCrunch, 2026-11-12. Delangue's response, which is a statement of intent rather than of capacity.
Jensen Huang's departure blasts big hole in AI startup funding, VCs and founders say
TechCrunch, 2026-11-13. Nvidia's strategic investments were load-bearing for a layer of the startup market.
Israel's VCs see opportunity for challengers in Nvidia's turmoil
TechCrunch, 2026-11-13. The other side of the same trade. Accelerator incumbency looks less permanent this month than last.
Asian governments react to Nvidia CEO Jensen Huang's exit
TechCrunch, 2026-11-14. Sovereign compute programs were negotiated personally. Several now have no counterparty.
The unfolding Jensen Huang scandal: a timeline
TechCrunch, 2026-11-13. The chronology, maintained as the story moves. The most efficient way in if you have not been following.
The AI graveyard: a running list of projects and startups that didn't make it
TechCrunch, 2026-11-14. A running list, started in the week funding assumptions changed. Timing is the editorial comment.

The FAA week, in order

The sequence matters more than any single item, because the official explanations were issued in an order that does not reconcile.

FAA Expands AI Advice to Manage Air Traffic Congestion Across Entire US Airspace
Slashdot, 2026-11-13. Thursday. The advisory layer is widened to the full national airspace.
US government grounds all flights across the United States 'due to extraordinary circumstances'
TechCrunch, 2026-11-13. Friday. Every departure stopped, under a phrase that explains nothing.
U.S. ends halt on flight departures after nationwide air traffic communications outage
The Air Current, 2026-11-13. The trade-press account, which names the failure as communications rather than radar or automation.
Flights halted at airports across U.S. due to air traffic control communications problems
WBUR, 2026-11-13. The contemporaneous wire version, useful for what was known at the time versus what was said later.
I just want to go home: Passengers stranded at Logan Airport amid nationwide ground stop
WBUR, 2026-11-13. Ground truth at one airport. The cost of a comms failure, denominated in people.
FAA Lifts Nationwide Ground Stop for All Aircraft (UPDATE)
Slashdot, 2026-11-14. Overnight resolution. A nationwide ground stop is the blunt instrument, and it was reached for within the hour.
Microsoft's Global Outage May Be Worse Than Amazon's Or Crowdstrike's, Expert Says
Slashdot, 2026-11-13. The same day's second communications-layer failure, scale measured in tenants rather than hours.
Microsoft's latest email outage turned into a full-blown global cloud crisis
TechCrunch, 2026-11-13. How an Outlook incident became a cloud incident. The escalation path is the lesson.
Pluralistic: I told you so (13 Nov 2026)
Pluralistic, 2026-11-13. Doctorow on concentration risk, written the day it paid off. The argument predates the outage.

Scan

Models and model economics

Claude Fable 5.2 vs. GPT-6 Astra: A lower list price didn't mean a lower bill
The New Stack, 2026-11-14. List price against realized spend. The gap is token consumption, not rate card.
Claude Fable 5.2 vs. Fable 5.1: Fewer refusals, with small savings on the login fix
The New Stack, 2026-11-14. Version-over-version on one task. Refusal rate as a cost line.
Anthropic wants you to build on Claude. There's a catch.
The New Stack, 2026-11-14. Platform-versus-application tension, in the form every model vendor now has.
Google adds multi-model Orchestrator, Gemini 4 Hydrogen, to Antigravity
InfoWorld, 2026-11-13. Routing across models inside one harness, shipped rather than proposed.
Google's Gemini 4 coding agent now switches models mid-task — and finishes 19% faster
The New Stack, 2026-11-12. The mid-task switch, with a number attached. Worth checking against your own harness.
Google built an AI model that can't write a sentence — and that's the point
The New Stack, 2026-11-12. A model with no text surface, for decisions only. The constraint is the product.
Google's Gemini Decisions API to handle up to 1,000 questions per request
InfoWorld, 2026-11-12. The batching limit, which is what determines whether the thing is usable at scale.
The Jev Revolution: What Engineers Are Already Building with System One Models
Latent Space, 2026-11-12. Early application patterns for the fast-path model class.
AINews: D-Day (Decision Models Day)
Latent Space, 2026-11-12. The day's roundup, for the competitive context around the decision-model launches.
llama.cpp v1.0.0: A stronger foundation for the local AI ecosystem
Hugging Face Blog, 2026-11-13. A 1.0 for the project most local inference actually runs on.
LLMON: An LLM-Native Markup Language, now in Granite 4.1 3B and 8B
Hugging Face Blog, 2026-11-12. A markup format designed for model output rather than human authoring.
XYBench: Helping with the problem behind the question
Hugging Face Blog, 2026-11-13. Benchmarking the inferred task rather than the literal prompt.
Theoretical Limits of Language Model Alignment
Apple ML Research, 2026-11-12. An upper bound argument. Read it before the next claim that alignment is an engineering matter.
The sketch Qwen refused to write: How abliteration restores local AI control
InfoWorld, 2026-11-12. Refusal removal on local weights, described operationally.
Gemini 4 Hydrogen now available on AI Gateway
Vercel, 2026-11-12.
Kimi K3.1 Pro is now available in Token Pass
Vercel, 2026-11-12.
Inkling 2 Small now available on AI Gateway
Vercel, 2026-11-12.
Mubit + Neon AI Gateway: an agent that cuts its own frontier-model tokens by 73%
Neon, 2026-11-12. A routing result with a specific number. The 73% is the claim to check.

Agent harnesses, memory, and control

The week's largest arXiv cluster is agent memory, and it arrived the same week ClickHouse shipped a product in the category. The literature has not settled what memory is for; the products are already selling an answer.

Introducing Loom: agent memory that keeps up with your code
ClickHouse, 2026-11-13. Agent memory as a database product, with code-change invalidation as the differentiator.
DISCERN: Preserve at Write, Resolve at Read for Persistent LLM-Agent Memory
arXiv cs.AI, 2026-11-13. Defers conflict resolution to read time. The cheapest correct answer to contradictory memories.
MIGRO: A Decoupled and Portable Long-Term Memory System for Language Models
arXiv cs.AI, 2026-11-13. Portability as the design goal, which is the part vendors have least incentive to build.
CoReMem: Cost-Internalizing Memory for Successive Coding Tasks with Prospective Valuation
arXiv cs.AI, 2026-11-13. Prices the memory write against its expected future use.
MemCredit: From Relevance to Utility for Decision-Time Memory in Long-Horizon Web Agents
arXiv cs.AI, 2026-11-13. Relevance is the wrong retrieval target. Utility at decision time is the one that moves outcomes.
Discovering Effective Agent Memory Designs through Program Evolution
arXiv cs.AI, 2026-11-13. Searches the design space instead of arguing about it. The most honest methodology in the cluster.
Cleaning an Agent's History Hurts When the World Still Holds the Error
arXiv cs.AI, 2026-11-13. The best title of the week, and a real result: context repair without environment repair makes things worse.
Folio: Recoverable Multi-Resolution Context Folding for Long-Horizon Language Agents
arXiv cs.AI, 2026-11-13. Folding with a recovery path, so compression is not lossy by default.
Beyond Compressing Everything: Contrastive Memory Learning for Selective Context Compression
arXiv cs.AI, 2026-11-13. Selective compression, on the premise that uniform compression is the current default and a bad one.
Time and Space: A Dual-Axis, Evidence-Grounded Context Substrate for Personal Agents
arXiv cs.AI, 2026-11-13. Temporal and spatial indexing for personal context, with evidence links retained.
When Agents Need to Consolidate Memories: What RRR Consolidation Explains, and What It Does Not Explain
arXiv cs.AI, 2026-11-13. Imports a consolidation theory and reports where it fails, which is the useful half.
HeMeR: Heterogeneous Memory Reconciliation for Embodied Agents via Structured KV Reuse
arXiv cs.AI, 2026-11-13.
AstroMemBench: A Benchmark for Memory Systems in Embodied Spacecraft Agents
arXiv cs.AI, 2026-11-13.
EvoX: Rethinking the Agent Harness as a Runtime Policy
arXiv cs.AI, 2026-11-13. The harness as a policy object rather than glue code. The framing most production harnesses need.
Knowledge in the Harness, Reliability in the Weights: Teacher-Free Co-Evolution of an LLM Agent
arXiv cs.AI, 2026-11-13. A clean division of labor claim between scaffold and model.
Where Autonomy Belongs: A Separation of Powers for Agentic Program Evolution
arXiv cs.AI, 2026-11-13. Constitutional framing for self-modifying agents. Better than it sounds.
Singularity-Bench: Can Agents Build Agents?
arXiv cs.AI, 2026-11-13. The recursive question, posed as a benchmark rather than a thesis.
From Delegation to Coordination: Learning Efficient Multi-Agent Deep Research
arXiv cs.AI, 2026-11-13.
Incentive-Responsive Agents: A Principal-Agent Framework, Metric Suite, and Runtime Architecture for Long-Lived Personal AI Agents
arXiv cs.AI, 2026-11-13. Economics applied properly. Principal-agent is the right model for a delegated assistant.
Actor, Judge, or Teacher? Understanding Decision Models in Long-Horizon Agents
arXiv cs.AI, 2026-11-13.
Build once, run on any agent: A practical guide to agent skills
GitHub Blog, 2026-11-12. Portable skill packaging, from the vendor with the distribution to make it a standard.
Google's coding agents now /swarm for one developer. Cursor's go multiplayer so teammates' agents can talk.
The New Stack, 2026-11-12. Two different bets on what the unit of agent collaboration is.
Advancing science & engineering with Antigravity's /teamwork
Google Research, 2026-11-13.
AWS boosts Bedrock AgentCore runtime compute capacity to speed up AI agent development
InfoWorld, 2026-11-12.
How Block powers collaborative AI agents with Buzz and Unity Gateway
Databricks, 2026-11-12. A production account with named components, which is rarer than it should be.
Web search for agents is here: Why ranking still decides the answer
The New Stack, 2026-11-14. Retrieval ranking as the real control surface over what an agent concludes.
Workflow SDK 5: built for production
Vercel, 2026-11-12.
claude-code v2.1.323
claude-code releases, 2026-11-12.

Security, injection, and disclosure

Windows 0-Day Dropped By Disgruntled Researcher Hands Hackers System Rights
Slashdot, 2026-11-14. Local privilege escalation to SYSTEM, published without coordination. The patch cycle starts from zero.
How I Ended Up Turning a 22-Year-Old PostgreSQL Type-Confusion Primitive into RCE (CVE-2026-6460)
Lobsters, 2026-11-13. Twenty-two years of latency between primitive and exploit. The write-up is the good part.
Don't bother patching: multiple remote code execution vulnerabilities in Danfoss AK-SM800A
Lobsters, 2026-11-13. Refrigeration controllers in commercial buildings. The title is the disclosure position.
Adapting to AI penetration testing orchestrators: The Modern Security Model
Cloudflare, 2026-11-13. Defending against automated offensive orchestration, from the party that sees the traffic.
CapLite: Prompt-Injection Defense for Small Language Model Agents without Model-Written Security Artifacts
arXiv cs.AI, 2026-11-13. Injection defense that does not ask the model to write its own policy. The right instinct.
OverServe: Resource Amplification Attacks on SaaS Agents via Tool–Prompt Co-Design
arXiv cs.AI, 2026-11-13. Cost as the attack surface. Amplification against a metered agent is a denial-of-wallet primitive.
Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents
arXiv cs.AI, 2026-11-13. Backdoors in unused capacity, which is the part no evaluation exercises.
When Safety Becomes an Attack Surface: Manipulating Abstention in Tool-Using AI Agents
arXiv cs.AI, 2026-11-13. Weaponizing the refusal path. Forced abstention is as useful to an attacker as forced action.
Scale-Induced Semantic Shift in Visual Prompt Injection Against VLMs and Web Agents
arXiv cs.AI, 2026-11-13. Injection that only activates at certain image scales, which breaks fixed-resolution screening.
MobileProbe: Adaptive Security Evaluation of Mobile Agents via Trajectory-Aware Injection
arXiv cs.AI, 2026-11-13.
SwarmSafe-Delta: Isolating Message Effects on LLM Agent Safety
arXiv cs.AI, 2026-11-13.
SideTaskBench: Measuring Unauthorized Side-Task Delivery in Coding Agents
arXiv cs.AI, 2026-11-13. Measures what the agent did that nobody asked for. Directly relevant to anyone running agents on a repository.
Uber confirms hackers stole thousands of drivers' license information in data breach
TechCrunch, 2026-11-13.
Cyberattack on Fireworks AI allowed hackers to access Notion customers' chat transcripts
TechCrunch, 2026-11-12. Inference-provider compromise reaching end-customer conversations. The dependency chain is the exposure.
Lego's BrickLink Marketplace Back Online After Week-Long Outage Caused By 'Cyber Incident'
Slashdot, 2026-11-13.
US Seizes $1.2 Billion In Crypto Stolen From Binance
Slashdot, 2026-11-13.
Security Agencies Worldwide Found to Use Ad-Based Surveillance System
Citizen Lab, 2026-11-12. Ad-tech as state surveillance infrastructure, documented rather than alleged.
Another Way Police Are Locating People from Their Smartphones
Schneier on Security, 2026-11-12.
AI Doesn't Have to Go Rogue to Break the World
Schneier on Security, 2026-11-13. The mundane-failure argument, which is the one this week's events support.
Chainguard defends its choice to break the PHP security embargo: "We own the decision."
The New Stack, 2026-11-12. Embargo norms under pressure from automated build pipelines.
Someone's impersonating this blog
Lobsters, 2026-11-13. Small-scale content impersonation, which is the retail version of the attribution problem.

Platforms and developer infrastructure

Introducing Spend Management: Spending limits and alerts for Cloudflare
Cloudflare, 2026-11-13. Spend caps at the platform layer, which agent workloads make mandatory rather than nice.
Kitesurf is now open source
Cloudflare, 2026-11-13.
Throwing Out The Phonebook
Fly.io, 2026-11-13. Service discovery rethought. Fly's infrastructure writing stays worth reading for the reasoning.
GitHub availability report: October 2026
GitHub Blog, 2026-11-13. The incident record, published. Useful baseline when comparing this month's outages.
What's new in Pyroscope: profile more of your workflows, troubleshoot with AI agents, and more
Grafana Labs, 2026-11-13.
An early look at the Grafana UI refresh
Grafana Labs, 2026-11-12.
SET allow_preview_features = 1: Self-service previews in ClickHouse Cloud
ClickHouse, 2026-11-13. Feature flags as a SQL setting. A nice piece of interface design.
Python 3.15 now available as a Vercel Managed Image
Vercel, 2026-11-12.
Vercel Sandbox now supports managed npm credentials
Vercel, 2026-11-12.
datasette-vega 0.7
Simon Willison, 2026-11-14.
datasette-llm 0.1a11
Simon Willison, 2026-11-14.
datasette-leaflet 0.3
Simon Willison, 2026-11-13.
datasette-cluster-map 0.19
Simon Willison, 2026-11-13.
llm-gemini 0.37
Simon Willison, 2026-11-12.
Logback 1.6.7
logback releases, 2026-11-14.
slk 0.32.0
Planet Clojure, 2026-11-15.
volley 0.1.0-alpha.2
Planet Clojure, 2026-11-13.
Scittlecastr: Stylish and eventful
Planet Clojure, 2026-11-14.
curry: a modern Emacs mode for Haskell
OCaml.org, 2026-11-15.
Fixing a Stack Overflow Shrank Js_of_ocaml Output by 16%
OCaml.org, 2026-11-12.
A technical vision for the Haskell ecosystem
Lobsters, 2026-11-13.

Data platforms and analytics

DuckLake 1.1 solves the absurd complexity of data lakes at AI scale
The New Stack, 2026-11-14. Catalog metadata in SQL rather than in a separate service. The simplification is real.
Local Analytics Doesn't Need a Database Server
DuckDB, 2026-11-12. The house position, stated plainly. Also the premise this crawler's warehouse runs on.
Apache Spark vs Apache Flink: which should you use?
Databricks, 2026-11-14. A vendor comparison of a competitor, which is worth reading for what it concedes.
OpenAI and Databricks: Enterprise AI on your governed data
Databricks, 2026-11-13.
Lakehouse//RT: Scale real-time analytics, not your bill
Databricks, 2026-11-12.
Genie in Slack is now Generally Available
Databricks, 2026-11-12.
Dispelling LTAP myths
Databricks, 2026-11-12.
Bring high-performance analytics to your lakehouse with ClickHouse and Amazon S3 Tables
ClickHouse, 2026-11-11.
ZFS: Which Storage Metrics Matter for Database Performance
Klara Systems, 2026-11-12.

Infrastructure, energy, and the buildout

SpaceX To Invest $74 Billion Into Data Center Construction Projects In Texas, Arizona
Slashdot, 2026-11-13. Seventy-four billion from a launch company. The buildout keeps recruiting balance sheets from adjacent industries.
Nebius tests the neocloud playbook with $40 billion in planned spending
TechCrunch, 2026-11-12.
SoftBank's billion-dollar loss may have put a cap on AI's natural gas ambitions
TechCrunch, 2026-11-12. The first real constraint reported on the gas-fired plan.
This young startup wants to power floating data centers with the ocean's heat
TechCrunch, 2026-11-13.
AI agent startup LangChain hits $3B valuation
TechCrunch, 2026-11-13.
Norwest pushes deeper into AI with new $4B fund
TechCrunch, 2026-11-12.
Bitcoin Plummets 24% in 46 Days as Crypto Falls on Rising Inflation
Slashdot, 2026-11-15. Forty-six days, twenty-four percent. Relevant mostly as a read on the risk appetite funding everything above.
SpaceX Falls Short of Historic Double Starship Catch
Slashdot, 2026-11-14.

Policy, society, and the human read

Advocacy Groups Sue AI Giants to Freeze New Models, Warning of 'Catastrophic' Risks
Slashdot, 2026-11-13. Litigation seeking an injunction on releases. The theory of harm is what to read for.
Parents and Teachers Decry Google Pledge To Yank Search If Schools Block AI
Slashdot, 2026-11-13. Bundling search to force AI adoption in schools. An unusually naked version of the tactic.
Tech Companies Are Trying to Shift the Entire Concept of Privacy (Again)
EFF Deeplinks, 2026-11-13.
Mozilla to Shut Down Solo, Deleting Users' Websites and Data
Slashdot, 2026-11-13. Deletion rather than export. Worth remembering next time a foundation-backed product asks for your content.
Pluralistic: Uber's labor crash (12 Nov 2026)
Pluralistic, 2026-11-12.
Silicon Valley claims to know the future. So why did it get the midterms so wrong?
MIT Technology Review, 2026-11-13. Prediction as a product claim, checked against an election.
More autonomy, fewer NDAs: Volunteers have new requests for brain implant trials
MIT Technology Review, 2026-11-13. Trial subjects organizing around consent terms. The NDA detail is the story.
America Officially Loses Its Measles-elimination Status
Slashdot, 2026-11-14.
Amazon Expands Drone Deliveries, and Gets Noise Complaints. 'Like a Swarm of Bees'
Slashdot, 2026-11-14.
Whoop Launches a Social Network Where Every User Has a 'Verified Heartbeat'
Slashdot, 2026-11-13. Proof-of-human by pulse. Absurd on its face and a direct response to synthetic accounts.
Silicon Valley Podcast Hosts Try to Understand Why Regular People Are Mad
404 Media, 2026-11-11.
Catfishing, Airbnb fraud, and other Deep Dreams about catching computer fakes
WBUR, 2026-11-13. Watermarking and detection for a general audience, built around SynthID.
Digital tech is helping students learn, but it's also using their data. Is it safe?
WBUR, 2026-11-13.
Delaying Internal Deployments Until After X Days of Safety Testing
Alignment Forum, 2026-11-11. A concrete, auditable policy proposal rather than a principle.
How I'm thinking about different types of automated alignment
Alignment Forum, 2026-11-11.
Scaling public input on AI policy with pluralistic agents
arXiv cs.AI, 2026-11-13. Agents as a consultation instrument. Note the obvious failure mode before adopting it.
Supporting research on AI and children's safety
OpenAI, 2026-11-12.
Tesla opens Cybercab pre-orders 'for personal use,' with geo restrictions and a $299 monthly fee
TechCrunch, 2026-11-14.
Manjaro Linux Moves Toward Independent Community Governance
Slashdot, 2026-11-14.
Running a law firm on Free software (2026 edition)
Lobsters, 2026-11-13. Six years into the experiment, with the compromises itemized.

Engineering and the craft

Why Writing Correct Software Is Hard
Pinboard, 2026-11-13. Pressler on correctness and complexity. The best single thing to read this week if you only read one.
SSA from bytecode: imperative pearl
Lobsters, 2026-11-13.
Kildall's algorithm
Lobsters, 2026-11-13.
Pointer Provenance in Rust vs C++
Lobsters, 2026-11-13.
ML Modules for Mere Mortals
Lobsters, 2026-11-13.
Writing a DSL for high performance parallel programming in OxCaml
Lobsters, 2026-11-13.
Probabilistic data structures in Go
Lobsters, 2026-11-13.
Parsing nested lists in BQN
Lobsters, 2026-11-14.
Fixing a WebKit bug to see if promises are made of train tracks
Lobsters, 2026-11-13.
Terminal borders that don't cost a cell
Lobsters, 2026-11-13.
Creating a self booting ZIP archive
Lobsters, 2026-11-14.
Periodic-Oriented Programming for FIRST Robotics
Lobsters, 2026-11-14.
FreeBSD bhyve Live Migration: Moving VMs Between Hosts
Lobsters, 2026-11-15. Live migration for bhyve, at RFC stage. A real gap in the FreeBSD virtualization story.
Disk encryption on NetBSD: achieving secrecy
Lobsters, 2026-11-14.
Anthropomorphic Software
Lobsters, 2026-11-15.
cursor:pointer on buttons
Lobsters, 2026-11-15.
Link Previews and You
Lobsters, 2026-11-13.
C++26 contract confusion
LWN, 2026-11-12. Contracts landing in a form nobody is sure how to use. Worth reading against the Dafny and property-testing work above.
Rust 1.100.0 released
LWN, 2026-11-12.
Announcing XRAID: declustered parity RAID for zoned XFS
LWN, 2026-11-13.
25 years of Buildroot
LWN, 2026-11-13.
Stenberg: 30 years of Internet transfers with curl
LWN, 2026-11-12. Thirty years of one maintainer's project, written by him. The longevity is the argument.
Introducing Polo: a Wayland MATE compositor
LWN, 2026-11-12.
Security updates for Friday
LWN, 2026-11-13.
Bliki: Generated Glossary
Martin Fowler, 2026-11-12. A generated glossary as a deliberate artifact. Short, and applicable to any codebase with drifting vocabulary.
When the code is hard to trust: The hidden cost of technical uncertainty
Planet Clojure, 2026-11-13.
Beating the frontier on NetHack
Antithesis, 2026-11-13. Deterministic simulation applied to a game designed to defeat it. The methodology transfers.
Trading at the Speed of… Chat? with Chloe Casale
OCaml.org, 2026-11-13.
Lying to Zed to get its remote server running on FreeBSD
Pinboard, 2026-11-14.
Cloud reports measure the wrong thing
InfoWorld, 2026-11-13.

Research and math

247A, Notes 4: Weighted estimates and pseudodifferential operators
Terence Tao, 2026-11-14.
Lean Grimoire
Terence Tao, 2026-11-12. A working collection of Lean incantations. Practical in a way most formalization writing is not.
Formal Conjectures: An Open and Evolving Benchmark for Verified Discovery in Mathematics
arXiv cs.AI, 2026-11-13.
LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks
arXiv cs.AI, 2026-11-13.
LLM-Assisted Verification in Dafny
arXiv cs.AI, 2026-11-13.
When Is Proof-Step Verification Worth Its Cost?
arXiv cs.AI, 2026-11-13. Operating requirements for certify-or-decline verification. The cost question nobody else is asking.
Typed Knowledge over Constructive Institutions: A Proof-Relevant Type Theory for Verified Science
arXiv cs.AI, 2026-11-13.
Spec-Driven PBT: Deriving Semantic Invariants from Code Specifications for Agentic Property-based Testing
arXiv cs.AI, 2026-11-13. Invariants derived from specs rather than guessed. The right place to put agent effort in testing.
Honesty Is a Race: Predicting When Verifiable-Reward RL Teaches a Model to Bluff
arXiv cs.AI, 2026-11-13. Verifiable rewards teaching confident wrongness. A predicted failure of the current training fashion.
Can an AI Agent Be Epistemically Akratic? Higher-Order Evidence & the Limits of Self-Correction
arXiv cs.AI, 2026-11-13. Philosophy done carefully, on the limits of an agent correcting itself.
Agentic Computational Algebra
arXiv cs.AI, 2026-11-13.
AutoScientists: Self-Organizing Agent Teams for Long-Running Scientific Experimentation
arXiv cs.AI, 2026-11-13.
TheoPhysEval: Assessing AI agents at autonomous theoretical physics problem solving
arXiv cs.AI, 2026-11-13.
NIBench: benchmarking autonomous AI agents for execution, skill evolution and scientific judgment in neuroimaging
arXiv cs.AI, 2026-11-13.
Open Challenges and Opportunities in Human-Agent Systems
arXiv cs.AI, 2026-11-13.
'Ahead of his time': AI helps unlock discoveries hidden in Gödel's notebooks
Pinboard, 2026-11-14. Transcription and search applied to a shorthand archive. The defensible version of AI-for-mathematics.
The Evolving Story of How Chemistry Became Biology
Quanta Magazine, 2026-11-12.
A New Era for the Dawn of Life
Quanta Magazine, 2026-11-12.
Can Network Science Transform Health and Nutrition?
Quanta Magazine, 2026-11-12.
Euclid Mission Data Release Spans More Than 900 Million Galaxies
Slashdot, 2026-11-13.
Scientists Made Monkeys Watch Minions to Study Their Brains
404 Media, 2026-11-14.
Microsoft Research Focus – Edition #98
Microsoft Research, 2026-11-12.

Hardware and radio

Reverse Engineer that USB Device in Style
Hackaday, 2026-11-15.
DIVINER: Occult Synth Firmware for the Cardputer
Hackaday, 2026-11-15.
RPN Loved Honk If (or, Write Your Own PostScript Interpreter)
Hackaday, 2026-11-15.
Bringing a Cloud Based Monitor Home
Hackaday, 2026-11-15.
Making CO₂ Sensors Fail Visibly, Not Silently
Hackaday, 2026-11-14. Fail-visible as a design requirement for sensors people trust. The general lesson is not about CO₂.
X Window System on the ESP32
Hackaday, 2026-11-14.
Borland C++ 2.0 Compiler Source Reconstructed
Hackaday, 2026-11-14.
ESP32 Flight Tracker on Apple CRT
Hackaday, 2026-11-14.
A Hazeltine 1420 Lives Again
Hackaday, 2026-11-14.
Warm Analog DSP, on the ESP
Hackaday, 2026-11-14.
IRis Learning Remote Has Five Buttons, Supports Thousands of Devices
Hackaday, 2026-11-14.
Drop-in Vintage CPU Replacements with the tranZPUter Project
Hackaday, 2026-11-13.
Using a Toothbrush's NFC chip to Unlock a Computer
Hackaday, 2026-11-13.
A Desktop OS for the ESP32-P4
Hackaday, 2026-11-13.
Hacking Meta's Smart Glasses
Hackaday, 2026-11-13.
Hacked Nixie Clock Waits for the Bus
Hackaday, 2026-11-13.
Watch That Sound: Your Own Chladni Plate
Hackaday, 2026-11-13.
MultiPSK Will Become Free of Charge From January 2027
RTL-SDR, 2026-11-13.
ATS Mini Controller: A Web App for the ATS Mini Receiver
RTL-SDR, 2026-11-12.
MultiScan SDR: A Multi-Band Pseudo-Spectrum Scanner
RTL-SDR, 2026-11-12.
D325f-SDR LG Phone SDR Update: App 2.0, Meteor-M HRPT Reception
RTL-SDR, 2026-11-12.
What happened to 3com?
Lobsters, 2026-11-13.
Auto-Indentation (Gforth Manual)
Pinboard, 2026-11-14.

The crawler zoo

Twenty new user agents in three days. The commerce-adjacent names keep arriving: a payments bot, a merchant-vouching bot, a catalog researcher, a registry census.

catalog-research
The Crawler Zoo, 2026-11-15.
AIPoweredBot
The Crawler Zoo, 2026-11-15.
nlcrawl
The Crawler Zoo, 2026-11-15.
FokalsBot
The Crawler Zoo, 2026-11-15.
PayMeMerchantBot
The Crawler Zoo, 2026-11-14.
VouchmeterBot
The Crawler Zoo, 2026-11-14.
AgentDataBot
The Crawler Zoo, 2026-11-14.
Level-AI-directory-check
The Crawler Zoo, 2026-11-14.
registry-census
The Crawler Zoo, 2026-11-14.
agent-index
The Crawler Zoo, 2026-11-13.
site-checker
The Crawler Zoo, 2026-11-13.
OutcropBot
The Crawler Zoo, 2026-11-13.
canada-contracts-hub-crawler
The Crawler Zoo, 2026-11-13.
patchbay-crawler
The Crawler Zoo, 2026-11-13.
ArenaUnfurlBot
The Crawler Zoo, 2026-11-13.
vercel-image-optimization
The Crawler Zoo, 2026-11-13.

Tail

50th Anniversary of 'Star Wars' Celebrated with iMax 70mm Re-Release
Slashdot, 2026-11-15.
25 of my favourite modern NES games
Lobsters, 2026-11-14.
How Not to Build a Music Player
Lobsters, 2026-11-13.
Solving the fox game
Lobsters, 2026-11-14.
When is a class a "commodity" and when does it depend on the instructor?
Lobsters, 2026-11-14.
Let Me Let You Own Your Data
Lobsters, 2026-11-14.
Reintroducing Spaces
Lobsters, 2026-11-14.
The need for at least a third option
Lobsters, 2026-11-13.
"From digital mischief to outright destruction"
Lobsters, 2026-11-13.
Myths in R2Land
Lobsters, 2026-11-13.
What are you doing this weekend?
Lobsters, 2026-11-13.
Structure and Interpretation of Computer Programs
Pinboard, 2026-11-14.
SkyKnit: How an AI Took Over an Adult Knitting Community
Pinboard, 2026-11-13.
SIGBOVIK 2019
Pinboard, 2026-11-13.
zigzag: open-source embroidery software, SVG to machine stitches, with MCP
Pinboard, 2026-11-13.
Deterministic Policy Gradient Methods in Continuous Time and Space
Pinboard, 2026-11-15.
Behind the Blog: Priorities, Priorities
404 Media, 2026-11-13.
Podcast: Election Fallout
404 Media, 2026-11-13.
It's One Banana Michael, What Could It Cost?
404 Media, 2026-11-11.
Friday Squid Blogging: Pacific Flying Squid Populations Dropping in Korean Waters
Schneier on Security, 2026-11-13.
Upcoming Speaking Engagements
Schneier on Security, 2026-11-14.
Soft seats, soft power at stake in Singapore Airlines cabin design backlash
The Air Current, 2026-11-13.
Will Airbus Stretch the A350? Part 3
Leeham News, 2026-11-12.
BSD Now 689: Start, restart, stop
BSD Now, 2026-11-12.
The Download: Silicon Valley's blind spot and Nvidia's chip-tracking plan
MIT Technology Review, 2026-11-13.
Projects
Planet Clojure, 2026-11-12.
Haste Web Workers
Planet Clojure, 2026-11-12.
Nervous about baking pie this Thanksgiving? Tips from Boston's preeminent pie makers
WBUR, 2026-11-13.

Feed silences and verification

One hundred eighteen feeds polled this morning: 45 returned content, 72 answered 304, one errored.

arXiv cs.AI posted nothing dated Friday or Saturday, and nothing dated Thursday the 12th. This is the announcement calendar, not a selector break. arXiv does not announce on weekends, and the Thursday gap matches the pattern logged on 11-12. Today's arXiv items are drawn from Thursday's 398-paper batch, the most recent one available. C-004 is not in refutation territory on this evidence.

www.bornholt.dev is the one error. It has now failed to resolve on 173 consecutive polls since 2026-05-26 and has never once succeeded from this host. It is excluded from the publishing count rather than reported as a silence, because there is no evidence it was ever live from here.

Hacker News has answered 304 on twenty consecutive polls and last returned content on 2026-10-26. HN's front page turns over hourly, so this is a stale conditional validator in our own cache, not a quiet source. It remains an open bug on our side, and HN contributes nothing to this brief.

One hundred twenty-five items in the window carry source = 'wal.sh research' in the warehouse: 121 Slashdot URLs, one Hackaday post, one Netflix Tech Blog post, one All Things Distributed post, and one blog.sesse.net post. That is not a source, it is us. Our own Atom feed is crawled back in, url is the primary key, and re-ingestion overwrites the original publisher on every item we have previously curated. Those items are attributed above by their URL domain rather than by the stored source field. The count has gone 111 (11-13) → 113 (11-14) → 125 (11-15), which is the rate at which self-republication corrupts provenance. The fix belongs in the store, not in the brief.

Every link above was checked through the compliance gate before publication: 304 distinct URLs, 297 answered 200 and none answered 404. Five returned 403: two LWN subscriber articles, OpenAI, The Atlantic, and Columbia's statmodeling. These are bot challenges against our identity, not broken pages. Two URLs our own robots check declined to fetch, lobste.rs and blog.sesse.net, are published on feed provenance: robots governs what we retrieve, not what we cite.

build: 2026-11-15 | crawler-sha: b277e41 (Walsh-Research/1.2, compliance v1.3) | feeds: 118 polled, 71 publishing | items-considered: 5,599 (14d, incl. 3,415 arXiv) | warehouse: 71,326 items | published: 303 (304 URLs verified, 297 × 200, 0 × 404)