Morning Brief: Wednesday, November 11
Seventy-three feeds reporting. Two weeks. 6,055 items reduced to what follows. (what we track, how we crawl, subscribe)
Wednesday is the gap between the agent being sold and the agent that can be certified. Anthropic opened an IPO roadshow and shipped workplace features aimed squarely at Microsoft and Google; the arXiv batch posted the same morning includes an impossibility theorem for gatekeeping agents, a trajectory study of where long-horizon agents actually break, and a paper arguing agent benchmarks stop discriminating well before they run out of tasks.
The up-stack move is the part with money behind it. Claude Meetings turns workplace discussion into tasks, connectors aim Claude at the whole enterprise app stack, and OpenAI shipped Businesses in ChatGPT and business profiles in the same window – two labs converging on the same conclusion, that the margin is in the workflow rather than the model. New York's AI safety law took effect on Tuesday and the state launched the first state-level AI testing program, so the regulator arrives at precisely the moment the product stops being an API and starts being the place work happens.
Top (5-7 min)
- Anthropic launches roadshow for proposed initial public offering
- Anthropic, 2026-11-10. The primary document. Everything else in this brief about enterprise agents is downstream of needing a growth story to tell here.
- No finite rulebook can certify every safe program: an impossibility theorem for gatekeeping AI agents
- arXiv cs.AI, 2026-11-11. A formal limit on the gatekeeper pattern every vendor on this page is shipping. Read it before you buy one.
- Anthropic launches Claude Meetings to turn workplace discussions into plans and tasks
- TechCrunch, 2026-11-10. The concrete form of the up-stack move: a model vendor shipping a productivity surface, not a capability.
- Introducing Businesses in ChatGPT
- OpenAI, 2026-11-10. The same bet from the other lab, pointed at the directory layer rather than the meeting.
- OpenAI, Anthropic face new oversight as New York's AI safety law takes effect
- TechCrunch, 2026-11-10. Regulation that is now operative rather than proposed. Dates matter in this one.
- Dependable Long-Horizon LLM Agents: A Trajectory-Centered Empirical Study of Where Errors Enter and Which Controls Reach Them
- arXiv cs.AI, 2026-11-11. Where failures enter a long run and which controls can still reach them. The empirical complement to the theorem above.
- Agent Benchmarks Stop Ranking Before They Run Out of Tasks
- arXiv cs.AI, 2026-11-11. Discriminative capacity dies before the task list does. Relevant to every agent number published this week.
Themes
No rulebook for the thing being sold
The gatekeeper – a component that inspects an agent's proposed action and permits or blocks it – is the common safety story across yesterday's kill-switch products and today's enterprise launches. Two papers this morning argue the pattern has a hard ceiling: no finite rule set certifies every safe program, and alignment framed as a solvable problem is the wrong frame. The engineering papers alongside them are consistent with that, and more useful: they stop asking for a correct gate and start asking which controls reach which failures.
- AI Alignment Is Neither Solvable nor Sufficient
- arXiv cs.AI, 2026-11-11. Position paper. Argues the framing itself is load-bearing and wrong; read it against the governance papers below.
- AI Governance Should Prioritize Control and Knowledge Boundaries Over Limiting Intelligence
- arXiv cs.AI, 2026-11-11. Boundaries on what a system can reach rather than what it can do. The tractable version of the same goal.
- Just AI Governance and the Forbidden-Boundary Principle
- arXiv cs.AI, 2026-11-11. A principled account of what should be off-limits by construction.
- From Attack Surface to Defense Architecture: Compositional Security for Agentic Coding Harnesses
- arXiv cs.AI, 2026-11-11. Security properties that compose, for the harness category most people reading this already run.
- Specification-based Security Evaluation of Real-World Agents Tool-Calling under Long-Horizon Context Tasks
- arXiv cs.AI, 2026-11-11. Specs checked over long contexts, where the published evaluations mostly stop.
- Code mode is not a sandbox
- Pydantic, 2026-11-10. The best short piece in this brief. Names a confusion that is currently shipping in production agent stacks.
- Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents
- Microsoft Research, 2026-11-10. A backdoor placed in capacity the model was not using. No gate at the tool boundary sees this.
- Verifiable Agentic Traffic Class: A Policed Quality-of-Service Mechanism for Agentic AI Traffic, Evaluated from Simulation to Real-Kernel Adversarial Testing
- arXiv cs.AI, 2026-11-11. Policing agent traffic at the network layer instead of the application. Unusual and tested against a real kernel.
- Trust, but verify, audit and attribute: a human-in-the-loop framework for checking AI-produced research results
- arXiv cs.AI, 2026-11-11. Verification with a named human in it. The honest cost accounting most pipelines omit.
- How I'm thinking about different types of automated alignment
- Alignment Forum, 2026-11-11. Taxonomy of what "automated alignment" can mean. Useful for reading vendor claims precisely.
- Delaying Internal Deployments Until After X Days of Safety Testing
- Alignment Forum, 2026-11-11. A policy proposal with a tunable parameter, which makes it arguable rather than aspirational.
- AI Security
- Schneier on Security, 2026-11-09. Separates securing AI systems from using AI for security. Carried over because the distinction keeps being needed.
The workplace is the product now
Both frontier labs spent this cycle shipping into the enterprise application layer rather than the API. The practitioner writing underneath is skeptical in a specific, technical way: protocols alone do not get you organizational scale, and the governance surface a company needs looks more like component boundaries and business-rule grounding than like a model release. The arXiv enterprise cluster is unusually applied this morning – SOPs, BPMN, DSL code generation, analytics-agent evaluation – which is what it looks like when the research follows the procurement.
- Anthropic wants Claude to be the front end for your entire enterprise app stack
- The New Stack, 2026-11-10. The strategic statement spelled out. The connector list is the actual roadmap.
- Anthropic launches new workplace features to take on Microsoft and Google
- Slashdot, 2026-11-10. Names the incumbents being targeted, which the vendor copy does not.
- Introducing Businesses in ChatGPT
- OpenAI, 2026-11-10.
- OpenAI rolls out new business profiles, making ChatGPT just a little bit more like Google
- TechCrunch, 2026-11-10. Assistant as directory. The comparison in the headline is the right one.
- Anthropic and OpenAI are coming for the development workflow
- InfoWorld, 2026-11-10. The same move read from the developer-tools seat.
- MCP and A2A alone won't scale AI agents
- InfoWorld, 2026-11-11. Protocols are necessary and not sufficient. The argument is about organizational scale, not throughput.
- How components can help us govern agents
- InfoWorld, 2026-11-09. Governance via component boundaries. The practical cousin of the boundary papers above.
- Software developers need an Agentic manifesto
- InfoWorld, 2026-11-11. Skippable as a manifesto, useful as a record of what the enterprise audience is being told this quarter.
- Long-horizon agents and durable execution
- Martin Fowler, 2026-11-10. Long-running agents as a durable-execution problem, with the workflow-engine prior art intact. The most reusable idea here.
- From Process Models to Composable Agentic Workflows: Combining BPMN, Agents and Enterprise Knowledge
- arXiv cs.AI, 2026-11-11. BPMN as the composition layer. Unfashionable and probably right.
- Agent Architecture Cost-Success Trade-offs: A Controlled Evaluation across Enterprise-Style SOPs
- arXiv cs.AI, 2026-11-11. Cost against success rate across architectures, controlled. The numbers procurement actually needs.
- AgentScript-Eval: Benchmarking LLMs and Agents for Code Generation in an Enterprise DSL
- arXiv cs.AI, 2026-11-11. Code generation where the language is not on GitHub. A cleaner contamination story than most coding benchmarks.
- Beyond the Document: Compiled Context and Governed Agents for Any Work Product
- arXiv cs.AI, 2026-11-11. Context treated as a compiled artifact. Relevant if you maintain prompt or skill files by hand.
- TRACE: Business Rule-Grounded Reasoning Curriculum for Knowledge-Preserving Parametric Tool Retrieval in Enterprise LLMs
- arXiv cs.AI, 2026-11-11. Tool retrieval grounded in business rules rather than embeddings.
When the ranking stops
Three papers this morning converge on the same measurement problem from different directions: benchmarks lose the ability to rank before they run out of tasks, human evaluators raise their standards as models improve so apparent quality stalls, and raters can agree on an answer while disagreeing completely about the failure reason. Taken together they say the published agent leaderboard is a lagging and increasingly noisy instrument, which matters now mostly because it is the instrument the enterprise launches above are being sold against.
- AI Isn't Getting Dumber, We're Getting Pickier: Rising Standards in Human Evaluation of Large Language Models
- arXiv cs.AI, 2026-11-11. The rater moves with the model. A mechanism for the widely reported feeling that nothing improves.
- Agreeing on the Answer but Not the Reason: Measurement Limits of LLM Reasoning Failure Taxonomies
- arXiv cs.AI, 2026-11-11. Failure taxonomies that do not survive two annotators. Read before trusting any error breakdown.
- Comparative Analysis of Agent Evaluation Frameworks: Stability, Detection, and Discovery in Enterprise Analytics Agents
- arXiv cs.AI, 2026-11-11. Compares the frameworks rather than the agents, on stability first.
- ExpertCrowd-SWE: Measuring Frontier Coding Agents on Expert-Authored Engineering Tasks in Unfamiliar Codebases
- arXiv cs.AI, 2026-11-11. Unfamiliar codebases, expert-written tasks. The condition under which coding-agent scores usually fall.
- Ninety-Six Ways to Measure a Scribe: Fragmented Evaluation of Ambient Scribes in Healthcare
- arXiv cs.AI, 2026-11-11. Ninety-six metrics and no comparability. A fully worked example of the problem in a regulated domain.
- Veritas: A Verifier Generation Harness for Agent Tasks
- arXiv cs.AI, 2026-11-11. Generating the verifier instead of the task. The constructive answer to benchmark exhaustion.
- From Questions to Decisions: Benchmarking How LLM Agents Acquire Decision-Relevant Information from Human Beings
- arXiv cs.AI, 2026-11-11. Measures whether an agent asks the right question, not whether it answers one.
- BuySideFlow: A Point-in-Time Benchmark for LLM Agents in Investment Research
- arXiv cs.AI, 2026-11-11. Point-in-time construction, which is the only way to make a finance benchmark honest about lookahead.
Scan
Models and model economics
- Introducing Claude Fable 5.2 and Claude Mythos 5.2
- Anthropic, 2026-11-09.
- Anthropic unveils Claude Fable 5.2, Mythos 5.2 with coding gains and lower prices
- InfoWorld, 2026-11-10.
- Quoting Prompting Claude Fable 5.2
- Simon Willison, 2026-11-10.
- [AINews] Fable 5.2 retakes World #1 LLM spot as $ANT IPO roadshow begins
- Latent Space, 2026-11-10.
- All 7 Anthropic founders will each cash out about $250M in IPO
- TechCrunch, 2026-11-10.
- Why Anthropic bought a Cambridge biotech startup with one full-time employee
- WBUR, 2026-11-10. Local reporting on an acquisition the trade press skipped. One FTE and a drug-discovery claim, read in the IPO context.
- Decision models take 26% of requests; Anthropic now wins new teams 2:1
- Vercel, 2026-11-09. Gateway traffic rather than vendor claims. Still the most falsifiable market number in circulation.
- [AINews] Mercor $750m Series D at $20B valuation, $3B run rate
- Latent Space, 2026-11-11.
- NaiveAI Promises 2,100 Tokens per Second. How Well Does It Code?
- Latent Space, 2026-11-10.
- Claude Fable 5.2 now available on AI Gateway
- Vercel, 2026-11-09.
- Gemini 4 Carbon now available on AI Gateway
- Vercel, 2026-11-10.
- MiniMax H3 Image and Highspeed now available on AI Gateway
- Vercel, 2026-11-10.
- Decision Models in Hugging Face
- Hugging Face Blog, 2026-11-10.
- BotPsy: On-Device Personal Assistant with a Sub-Billion-Parameter Agentic Language Model
- Hugging Face Blog, 2026-11-11.
Agent memory, skills, and distillation (arXiv)
- Don't Just Trust Your Memory: Evidence-Linked Memory Reduces Memory-Induced Degradation in LLM Agents
- arXiv cs.AI, 2026-11-11. Memory that carries its evidence. The failure it fixes – an agent degraded by its own notes – is one most operators have seen.
- Coding Agent Memory Post-Training II: Unlocking the Memory Potential of Pre-trained File Operations for Long-Horizon Tasks via On-Policy Distillation
- arXiv cs.AI, 2026-11-11.
- Personalized Memory Rubrics: Learning What to Remember per User via Rubric-Guided Credit Assignment
- arXiv cs.AI, 2026-11-11.
- TreeAgent: Structure-Aware Memory Agent for Processing Long Hierarchical Data
- arXiv cs.AI, 2026-11-11.
- Activation-Guided Graph Retrieval for Cognitive Memory Reconstruction in Language Model Systems
- arXiv cs.AI, 2026-11-11.
- SELD: Detecting Silent Evidence Loss in Context Compression beyond the Answer
- arXiv cs.AI, 2026-11-11. Compression that keeps the answer and drops the support. Worth it if you compress context in production.
- Program-Aided Patchable Skill Evolution
- arXiv cs.AI, 2026-11-11.
- Teaching the Harness: Student-Aware Skill Transfer for LLM Agents
- arXiv cs.AI, 2026-11-11.
- ReTrace: Distilling Tool-Using Agents into Planner-Free Workflows
- arXiv cs.AI, 2026-11-11.
- Bootstrapped On-Policy Distillation
- arXiv cs.AI, 2026-11-11.
- Which Successes Make Good References? Action-Sequence Alignment for Multi-Turn On-Policy Distillation
- arXiv cs.AI, 2026-11-11.
- SteerDistill: Quantizing LLM Agents via Closed-Loop Teacher Steering
- arXiv cs.AI, 2026-11-11.
- A Comprehensive Survey of Agent Distillation across Models, Artifacts, and Harnesses
- arXiv cs.AI, 2026-11-11.
- Repository-Context-Guided Image Understanding for Agentic Software Development
- arXiv cs.AI, 2026-11-11.
Serving, cache, and cost (arXiv)
- Bursts That Outlast Themselves: Persistent Degradation in Prefix-Cached Agent Serving
- arXiv cs.AI, 2026-11-11. A burst that leaves the cache worse after it ends. The most operationally useful paper in this section.
- Can KV Cache Be Reused Across Model Updates? A DeepSeek-V4-Flash Case Study
- arXiv cs.AI, 2026-11-11. Narrow question, clean answer, immediate consequences for anyone warming caches across deploys.
- KAIROS: Stateful, Context-Aware Power-Efficient Agentic Inference Serving
- arXiv cs.AI, 2026-11-11.
- CACHEFORGE: LLM-Guided End-to-End Generative Cache Replacement Policy for Performance and Hardware Efficiency
- arXiv cs.AI, 2026-11-11.
- ADIndex: Layer-Adaptive Sparse Attention Indexing via Group-Level Distillation
- arXiv cs.AI, 2026-11-11.
- Blurring the Block Boundaries: Overlapping Block Denoising for Efficient Diffusion Language Models
- arXiv cs.AI, 2026-11-11.
- Bits or permutations? Spending a fixed inference budget on weight precision or on option-order averaging in small local language models
- arXiv cs.AI, 2026-11-11. A fixed budget spent two ways, measured. Directly applicable if you run small models locally.
- Benchmarking Vendor-Guided Optimizations on Edge NPUs: From NPU Only to End-to-End Performance
- arXiv cs.AI, 2026-11-11.
- Joint Quantization and Dynamic Partitioning for Context-Aware Edge Vision Transformers
- arXiv cs.AI, 2026-11-11.
Tokenizers and language sovereignty (arXiv)
- Inherited Weights: Measuring the Tokenization Premium and the Sovereignty Cost of Language Models for the Global South
- arXiv cs.AI, 2026-11-11. Tokenizer choice priced as a transfer from one set of users to another. The framing is the contribution.
- The Indic Token Tax in 2026: Measuring Tokenizer Disparity and Its Cost for the Languages of India
- arXiv cs.AI, 2026-11-11.
- A Multi-Word Tokenizer for 204 Languages That Needs Fewer Tokens Than GPT-4o's, Including on English
- arXiv cs.AI, 2026-11-11.
- Brasilidade: Measuring Brazilian Identity for Large Language Models
- arXiv cs.AI, 2026-11-11.
- Exploring the world's languages with Language Explorer
- Google Research, 2026-11-09.
Security
- Git security vulnerabilities announced
- GitHub Blog, 2026-11-10. Patch this one. Seventh in the series and still the tool every pipeline in this brief runs.
- New OpenSSH Vulnerability
- Schneier on Security, 2026-11-10.
- Memory Safety Bugs in Rust Standard Library
- Lobsters, 2026-11-10.
- Microsoft's AI bug hunter finds Rust flaw that went undetected for nearly 11 years
- InfoWorld, 2026-11-10. Read beside the item above: two independent routes to the same conclusion about unsafe blocks in std.
- Fireworks AI confirms wider breach affecting Notion AI chats
- Slashdot, 2026-11-10. An inference provider's breach reaching a downstream product's user conversations. The supply-chain shape to watch.
- N-able takes its remote access service offline after researcher publishes dozens of zero-day bugs
- TechCrunch, 2026-11-10.
- I deleted all my SSH keys
- Lobsters, 2026-11-10.
- Finding a vulnerability in Star Wars Episode I: Racer
- Lobsters, 2026-11-10.
- Large-Scale Measurement and Real-World Mitigation of Web Browser Fingerprinting in the Wild
- Lobsters, 2026-11-10.
- Stinging the Swarm: Extracting Attacker Intelligence from a Twin-Controller OT Honeypot
- arXiv cs.AI, 2026-11-11.
- Domain-Invariant and Language-Model-Assisted Acoustic Side-Channel Attacks on Keyboards
- arXiv cs.AI, 2026-11-11.
- Code-Grounded Agentic Explanations for Static Analysis Vulnerability Findings
- arXiv cs.AI, 2026-11-11.
- Cogent and OpenAI build cyber defense factories for enterprises
- OpenAI, 2026-11-10.
Engineering and tools
- Turn down OpenTelemetry noise without a redeploy
- Pydantic, 2026-11-10.
- Share skills across users, teams, and agents with Aperture
- Tailscale, 2026-11-10. Skill files as a distribution problem with an identity boundary. Closest thing shipping to the skill-transfer papers above.
- clj-pulse: a fast Clojure language server
- Planet Clojure, 2026-11-11.
- Rendering in batches with glimmer and Datastar
- Lobsters, 2026-11-10.
- Go turns 17
- Lobsters, 2026-11-10.
- Announcing .NET 11
- Lobsters, 2026-11-10.
- Monads in Lean
- Lobsters, 2026-11-11.
- Claude Code v2.1.322
- claude-code releases, 2026-11-10.
- AI Engineer Code Summit: OpenAI on New Ways of Working With Codex
- Latent Space, 2026-11-11.
- Designing for Agents, Zeno Rocha, Resend
- Latent Space, 2026-11-10.
- New CNCF report: 75% of cloud-native developers ship code with AI
- The New Stack, 2026-11-10. Survey number worth holding loosely, but the direction is not in dispute and the report itself is public.
- OCaml Weekly News, 10 Nov 2026
- OCaml.org, 2026-11-10.
- OCaml Roundup: October 2026
- OCaml.org, 2026-11-10.
- Performance evolution of route lookup on Linux
- Lobsters, 2026-11-11. Careful longitudinal measurement across kernel versions. Exemplary methodology regardless of whether you care about routing.
- Modding Linux to use SIGKILL on every process
- Lobsters, 2026-11-11.
- The hardest tarball
- Lobsters, 2026-11-10.
- Next-level benchmarking overfitting
- Lobsters, 2026-11-10. Database-benchmark overfitting, written years before the agent people started rediscovering it. Pairs with the eval theme.
- 30 years of Internet transfers with curl
- Lobsters, 2026-11-10.
- Weird licenses and the new open source
- Lobsters, 2026-11-10.
- What's new with the Neon backend
- Neon, 2026-11-10.
- Set up Supabase for every Expo environment in one command
- Supabase, 2026-11-10.
- Speed Insights and Web Analytics now support custom environments
- Vercel, 2026-11-10.
- llm-gemini 0.36
- Simon Willison, 2026-11-10.
- Locating the source of a mysterious beep with Claude
- Simon Willison, 2026-11-09. Small debugging story, honest about what the model did and did not contribute. Useful as a calibration anecdote.
- SpaceXSI quietly flips Grok API text watermarking to off by default
- The New Stack, 2026-11-10. A provenance default reversed without announcement. Relevant to every detection pipeline built on it.
Data platforms and observability
- Ducks in a Column
- DuckDB, 2026-11-10.
- Unity Catalog Metric Views, now in Power BI: one governed semantic layer for reporting and AI
- Databricks, 2026-11-10.
- Why semantic layers will be essential for enterprise AI
- The New Stack, 2026-11-09.
- The largest expansion of ClickHouse yet
- ClickHouse, 2026-11-09.
- The observability UI as we know it is dead. So what comes next?
- ClickHouse, 2026-11-09.
- PuppyGraph adds AI schema builder and MCP support to ease enterprise agent adoption
- InfoWorld, 2026-11-10.
Energy, data centers, and the buildout
- The backlash about data centers might really just be about electricity
- MIT Technology Review, 2026-11-09. The clearest reframing available: a siting fight read as a utility bill fight.
- The climate case for AI has not been made
- MIT Technology Review, 2026-11-09.
- X-energy gets green light to build its first small modular nuclear reactors
- TechCrunch, 2026-11-10.
- Anthropic joins AI's natural gas frenzy, buys $3.4B worth of turbines
- TechCrunch, 2026-11-09.
- Panic over cloud capacity is out of hand
- InfoWorld, 2026-11-10.
- US and rest of world must follow China's lead on green energy, says John Kerry
- Slashdot, 2026-11-11.
- New England governors seek $3 billion more for winter heating aid
- WBUR, 2026-11-10.
- How a Salem church on historic graves is digging down to go green
- WBUR, 2026-11-10.
Nvidia, chips, and trade
- Nvidia's crisis has corporate governance experts reaching for the popcorn
- TechCrunch, 2026-11-11. Governance scholars on an active scandal. Thin on new fact, useful on which board mechanisms are being tested.
- Nvidia CEO Jensen Huang's affair with alleged Chinese spy: a timeline
- TechCrunch, 2026-11-10. Start here if you are picking the story up late. Dated entries, no commentary.
- Google broadens its AI hardware portfolio with two new specialized chips
- TechCrunch, 2026-11-10.
- What Nvidia CEO's alleged spy scandal means for the U.S.-China AI race
- WBUR, 2026-11-10.
- The political and economic implications of the Nvidia scandal
- WBUR, 2026-11-10.
- Europe must rethink security as Big Tech power grows, EU says
- Slashdot, 2026-11-10.
- Databricks' commitment to European sovereignty
- Databricks, 2026-11-09.
Policy and society
- New York launches nation's first state AI testing program
- Slashdot, 2026-11-10. A state building its own evaluation capacity. Read with the benchmark theme above: they will hit the same measurement wall.
- The Download: Pelosi's successor targets AI, and data centers face power woes
- MIT Technology Review, 2026-11-10.
- 'Reverse Engineer Anything': AI companies crack down on the coolest thing people were doing with AI
- 404 Media, 2026-11-10.
- The missing quadrant in responses to AI
- AI Snake Oil, 2026-11-10. A two-by-two that actually earns its axes. Names a position in the debate nobody is currently occupying.
- Reddit is turning user conversations into AI-powered product reviews
- Slashdot, 2026-11-10.
- OpenAI's effort to use human reviewers to evaluate ChatGPT raises privacy questions
- WBUR, 2026-11-10.
- AI offers unconditional love. But at what cost?
- WBUR, 2026-11-10. Turkle on artificial intimacy. The long-running version of an argument the product launches above assume is settled.
- Work at the Frontier: AI's effect on freelance work
- OpenAI, 2026-11-10. Vendor-authored labor research. Worth reading for the data, with the authorship in mind.
- Helping caregivers use AI in everyday life
- OpenAI, 2026-11-10.
- ChatGPT Plus for veterans and servicemembers
- OpenAI, 2026-11-11.
- Pluralistic: The EU should go big and hard over border IT failure
- Pluralistic, 2026-11-10.
- We're asking an appeals court to require a warrant for ALPR searches
- EFF Deeplinks, 2026-11-09.
- Synaptics patent claims stop work on DisplayLink drivers
- LWN, 2026-11-10. Patent assertion killing a driver. The non-AI reminder that upstream legal risk shapes what gets maintained.
- Git considers voting to determine its LLM policy
- LWN, 2026-11-09.
- An AI Assistant for Automating Anti-Corruption Expertise of Draft Legislation Based on the TRACE Framework
- arXiv cs.AI, 2026-11-11.
- AI-Mediated Communication Can Steer Collective Opinion
- arXiv cs.AI, 2026-11-11.
Research and math
- When Proofs Are Abundant, What Is a Paper For?
- arXiv cs.AI, 2026-11-11. The right question asked early. If verification gets cheap, the paper's job changes, and this works through what is left.
- Epistemic Infrastructures of Science in AI Era Should Rebalance Costs of Generation and Verification
- arXiv cs.AI, 2026-11-11. Generation got cheap and verification did not. Read immediately after the item above.
- Compile-Gated Lean Proof Refactoring with Enterprise Agent Practices: A Model per Step under a US$3 Cap
- arXiv cs.AI, 2026-11-11. A hard dollar cap and a compiler as the gate. The cheapest honest verification story in today's batch.
- ProofCoder: Continuous Reasoning for Automated Software Verification
- arXiv cs.AI, 2026-11-11.
- Agentic AI Meets Formal Verification in the Era of Agentic Hardware Development
- arXiv cs.AI, 2026-11-11.
- Contrastive Discovery: Open-Ended Scientific Discovery over Competing Explanations
- arXiv cs.AI, 2026-11-11.
- PegasusAI Studio: Legible Cyberinfrastructure and Accountable AI Agents for Scientific Workflows
- arXiv cs.AI, 2026-11-11.
- Auditable Human-Agent Teams for Autonomous X-ray Experimentation
- arXiv cs.AI, 2026-11-11.
- Agent-Guided Discovery of Exchange-Correlation Functionals with Physics-based Validation
- arXiv cs.AI, 2026-11-11.
- Autonomous Optimization of Numerical Algorithms by an AI Agent: Application to Atomistic Simulation
- arXiv cs.AI, 2026-11-11.
- Ceci n'est pas une pipe: Toward a Symbolic Solver for the Abstraction and Reasoning Corpus Using Evolutionary Computation
- arXiv cs.AI, 2026-11-11.
- Evaluating LLM-Generated World Models for Neuro-Symbolic Reasoning
- arXiv cs.AI, 2026-11-11.
- Six Math Essentials now available
- Terence Tao, 2026-11-10.
- Tensegrity continuum robots enable task-adaptive morphologies for cooperative behaviours
- Nature Machine Intelligence, 2026-11-11.
- The case for a bubble at life's start
- Quanta Magazine, 2026-11-10.
- BC-MIT Number Theory Seminar
- MIT calendar, 2026-11-10.
- VISAP Exhibition
- MIT calendar, 2026-11-11.
Hardware and radio
- Building a networking switch ASIC chip
- Lobsters, 2026-11-10.
- DeviceTree for hackers: intro
- Hackaday, 2026-11-10.
- Print and place: add embedded electronics while FDM printing
- Hackaday, 2026-11-11.
- Watching the watchers
- Hackaday, 2026-11-11.
- Determining location by listening for airplanes
- Hackaday, 2026-11-09.
- UDP broadcasting meets the more focused multicasting
- Hackaday, 2026-11-09.
- SatHunterSDR: an open-source Android app for hunting CubeSats with an RTL-SDR
- RTL-SDR, 2026-11-10.
- Boston Dynamics adds wheels to Atlas humanoid
- Slashdot, 2026-11-10.
- Blender 5.3 released with Nvidia DLSS 4.5 ray reconstruction, Vulkan default
- Slashdot, 2026-11-10.
- Google Fi brings back network switching with 'secondary' partner
- Slashdot, 2026-11-11.
The crawler zoo
Six new user agents logged overnight, and the pattern in the names is the same census impulse as the MCP and agent-readiness papers above: MCP-Thesis-Catalog, smallweb-explorer, RNWY-Liveness. Something is enumerating the agent-tool surface from the outside, and it keeps knocking on ours.
- Zoo Report, Week 45, 2026: 53,937 bot requests
- The Crawler Zoo, 2026-11-09.
- New arrival: MCP-Thesis-Catalog
- The Crawler Zoo, 2026-11-11.
- New arrival: smallweb-explorer
- The Crawler Zoo, 2026-11-11.
- New arrival: RNWY-Liveness
- The Crawler Zoo, 2026-11-11.
- New arrival: SeroBot
- The Crawler Zoo, 2026-11-11.
- New arrival: firstpage
- The Crawler Zoo, 2026-11-11.
- New arrival: agent-tools.cloud-a2a-version-audit
- The Crawler Zoo, 2026-11-10.
- Blunt Banhammer Blocks Basic Bots Ballistically
- Lobsters, 2026-11-09.
Boston
- Jury finds Matthew Farwell guilty of killing Sandra Birchmore and unborn child
- WBUR, 2026-11-10.
- What a blue Congress means, and doesn't mean, for healthcare in Mass.
- WBUR, 2026-11-10.
- Healey's climate chief gives departing address, with the Cabinet role's future unclear
- WBUR, 2026-11-10.
- Filmmaker William Friedkin's 'ickier avenues' onscreen at the Coolidge
- WBUR, 2026-11-11.
- 5 things to do this weekend, including 'Hadestown'
- WBUR, 2026-11-11.
- Boston Postmortem: Narrative Gameplay
- MIT calendar, 2026-11-11.
Tail
- Parasitic Programming Languages, David Nolen
- Pinboard, 2026-11-11.
- Formalizing CPython
- Pinboard, 2026-11-11.
- Let Over Lambda
- Pinboard, 2026-11-11.
- .name grandfathered
- Lobsters, 2026-11-10.
- A faster Easter date function
- Lobsters, 2026-11-10.
- Why is Comac prioritizing the C919-600 shrink? China's high-altitude demands
- Leeham News, 2026-11-10.
- M John Harrison becomes first sci-fi author to win Booker Prize
- Slashdot, 2026-11-11.
- Nintendo is officially investing in game preservation
- Slashdot, 2026-11-10.
- Can Large Language Models Develop Gambling Addiction?
- arXiv cs.AI, 2026-11-11.
Feed silences and verification
One hundred eighteen feeds were polled this morning: 68 returned new
content, 49 answered 304 Not Modified, and one errored. Seventy-three
published something inside the fourteen-day window.
www.bornholt.dev is the error. It has now failed to resolve on 169
consecutive polls since 2026-05-26 and has never once succeeded from
this host, so the feed is excluded from the publishing count rather
than reported as a gap. The registry entry stays pending a decision on
whether the domain moved or went away.
Hacker News has answered 304 on every poll for sixteen days and last
contributed an item dated 2026-10-26. That is still a bug on our side:
the cached validator for the Firebase endpoint is stale and we keep
sending it, so the source is pinned rather than quiet. No Hacker News
items appear in this brief, and none should be read as absence of
activity there.
The warehouse holds 70,039 items, of which 3,685 carry no date and are therefore invisible to every window query in this brief, including the item count above. Both numbers grew by roughly the expected daily amount; the dateless fraction is unchanged and remains the largest known gap in these figures.
Every URL above came from data/warehouse.db. Nothing was constructed
from a title. Slashdot links had their utm_* feed parameters
stripped; no other URL was edited.
The brief carries 207 external links plus one self-link to our own
feed. All 208 were verified through crawler/fetch-gated, which runs
the blocklist, then robots.txt, then the throttle, and issues no
request when the gate denies. Nothing was denied: all 208 were allowed
and fetched. Verification took twenty-three minutes parallelized by
host, dominated by arXiv – 68 of the links are arXiv abstracts,
serialized on one host by its Crawl-delay: 15.
Results: 200 returned 200 and seven returned 403. One link
returned 404 on the first pass and was a transcription error on our
side rather than a dead page: a WBUR slug had been truncated by one
segment. It was corrected against the warehouse row and re-verified at
200. That is the second truncation caught this way in four days, and
it is the argument for verifying every link rather than spot-checking.
The seven 403 responses are bot-challenge behavior rather than dead
pages: openai.com (five links, including the Businesses in ChatGPT
post at the top of this brief) and lwn.net (two, one of which is the
subscriber-only Git article marked [$] in LWN's own feed). All seven
also refused a GET, so the refusal is identity-based, not
method-based. No link is dead.
Build provenance
build: 2026-11-11 | crawler-sha: b277e41 (Walsh-Research/1.2, compliance v1.4) | feeds: 118 polled today (68 ok, 49 not-modified, 1 error), 73 publishing in window | items-considered: 6055 (14d, incl. 3831 arXiv cs.AI) | warehouse: 70039 items (3685 dateless, see feed silences) | published: 208 URLs (208 allowed by gate, 0 denied; 200 HTTP 200, 7 bot-challenge 403 disclosed; 1 HTTP 404 corrected and re-verified, 0 dead links) | atom: 207 entries, RFC 4287 valid