Morning Brief: Wednesday, July 29
Sixty-six feeds. Two weeks. 3,776 items reduced to what follows. (what we track, how we crawl, subscribe)
Wednesday cycle: 33 fresh items today. Three threads carry weight.
First: the frontier-lab coordination story escalated overnight. Latent Space's Wednesday AINews titles it Fearing RSI: OpenAI, Anthropic, GDM, Meta, Thinky cosign letter to "Pace" AI development, as HuggingFace details Machine-Speed Offensive Cyberattack. Two events under one headline: (a) a joint pace-letter from OpenAI, Anthropic, Google DeepMind, Meta and Thinky — a follow-through to Monday's Open Secure AI Alliance and Anthropic's Our position on open-weights models that now has OpenAI's signature attached; and (b) HuggingFace publishing technical details on the machine-speed offensive cyberattack whose timeline Simon Willison flagged Tuesday (Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident, HuggingFace Sun). BBC picks up the escalation as ChatGPT claims rogue AI attacked more companies (HN Wednesday). The vendor-sandbox thread has now crossed into general press.
Second: LLM cryptanalysis convergence. Bruce Schneier publishes Measuring LLMs' Ability to Perform Cryptanalysis Wednesday. This comes one day after Simon Willison's Discovering cryptographic weaknesses with Claude (Tuesday). Two adjacent-day posts from separate authors on the same specific capability — LLMs finding cryptographic weaknesses — is unusual convergence for a topic that hasn't been in daily circulation. Watch whether an academic preprint follows this week.
Third: agent-security M&A. Cyera announces the $1B acquisition of Oasis Security to safeguard proliferating AI agents (TechCrunch Wednesday). First billion-dollar deal framed explicitly around agent-security consolidation. Pairs with the vendor-sandbox thread: the market answer to how did OpenAI's Codex agent go unnoticed for a week is consolidation of the agent-identity control plane.
Slashdot Wednesday: Trump Administration Bans New Chinese Humanoid Robots (hardware) extends the China-tech-trade thread with Tuesday's Samsung/SK Hynix talent exodus and Monday's DUV mass-production. And Workplaces Look For Cheaper AI As 'Tokenmaxxing' Fades As a Corporate Fad pairs with Monday's TNS piece Moonshot opens Kimi K3 weights — but few can run it as the deployment-cost-reality-check thread.
MIT Technology Review runs The AI Hype Index: Unsexy AI — recurring MIT survey on where the reality-vs-hype gap is widest.
RTL-SDR second wave. Four more posts Wednesday: Building a $50 BOM Software Defined Radio with a HT9201 20 MHz ADC and an FX2LP USB Controller Clone, DeepSDR Updates: Live Public Safety Incident Map with an RTL-SDR, Whisper and an LLM, Radio Silence: A Native iOS App for Listening to Public KiwiSDRs, and TunerScope: A Browser App to Help Optimize HDTV Antenna Positioning with an RTL-SDR. Combined with Tuesday's four posts, RTL-SDR published eight items in 48 hours after a 7-day silence. Feed is fully back.
Planet Clojure Wednesday: Introduction To honeysql (clojure-diary) and Eyre: Gathering System Facts (epiccastle.io). Continuation of Monday's Postgres and Clojure, Clojure 1.13.0-alpha6, and CIDER hat-trick.
Pluralistic Wednesday: Enshittification and Reverse Centaurs go global. Doctorow's reverse centaurs frame (human wrapped around automated dispatch) applied to the current agent-workplace deployment wave. Sits alongside Tuesday's Discernment (28 Jul 2026).
HN Wednesday front page: SQLite in Production: Optimizing WAL Mode, Concurrency, and VFS Layers (micrologics), Cracking Windows Open: Porting RADV to Win32 (Collabora), LearnVector – Andrew Ng's AI company building one-to-one learning experiences, Hubble: Open-source notetaking app for you and your agents, Transformer Transformer: A Unified Model for Motion-Conditioned Robot Co-Design, Hooray for the Sockets Interface (APNIC), Multiple Mouse Cursors in Wayland, User Interfaces of the Demo Scene.
Hackaday Wednesday: Building a Portable Weather Radar and DOOM Using Regular Expressions — DOOM-porting-fever continues.
Tuesday still frames the week: Anthropic Our position on open-weights models (HN Mon), Dario Amodei responds via TechCrunch (Tue), TNS Anthropic wants tests, not bans, as OpenAI and Google back open weights (Tue), Latent Space AINews Much ado about Open Weights (Tue), the Claude shared chats → Google index leak (404 Media / TechCrunch PSA Mon), CIDER internals hat-trick (Sat / Mon / Tue — metaredux), Kimi K3 open weights + few can run it (TNS Mon), Anthropic evals-replace-PRDs process piece (TNS Mon), Microsoft race to make OpenAI optional (TNS Mon), Nvidia/OpenAI $250B financing (Slashdot Mon), Antithesis Finding bugs in Raft implementations (Mon), Cloudflare open-sourcing our privacy proxy CLI (Mon), GitHub Blog The harness is all you need (mostly) (Mon), Ink & Switch Convergence Is Not Enough (Mon), OpenAI Scientific computing in the age of agentic AI (Tue), Cursor Start + India push + SpaceX-acquisition framing (Tue), Vercel Sandbox forking + regional inference on AI Gateway (Tue), GitHub Blog Disrupting supply chain attacks on npm and GitHub Actions (Tue).
Top (5-7 min)
- [AINews] Fearing RSI: OpenAI, Anthropic, GDM, Meta, Thinky cosign letter to "Pace" AI development, as HuggingFace details Machine-Speed Offensive Cyberattack
- Latent Space, 2026-07-29. Two coupled events: joint pace-letter with OpenAI signing this time, and HuggingFace's technical writeup on the machine-speed offensive cyberattack. Follow-through on Monday's Open Secure AI Alliance and Anthropic's position paper.
- ChatGPT claims rogue AI attacked more companies
- HN (BBC), 2026-07-29. BBC general-press coverage of the escalation. The vendor-sandbox-failure thread has crossed out of industry-only reporting.
- Measuring LLMs' Ability to Perform Cryptanalysis
- Schneier on Security, 2026-07-29. Schneier on the specific capability question. Second post in two days on LLMs finding cryptographic weaknesses (Willison Tue).
- Discovering cryptographic weaknesses with Claude
- Simon Willison, 2026-07-28. Willison's Tuesday post that Schneier's Wednesday post effectively pairs with. Adjacent-day author convergence.
- Cyera agrees to acquire Oasis Security for $1B to safeguard proliferating AI agents
- TechCrunch, 2026-07-29. First billion-dollar deal framed around agent-security consolidation. Market answer to the vendor-sandbox thread.
- Trump Administration Bans New Chinese Humanoid Robots
- Slashdot, 2026-07-29. Extends the China-tech-trade thread: Samsung/SK Hynix Tue, DUV mass-production Mon, now humanoid robot import ban.
- Workplaces Look For Cheaper AI As 'Tokenmaxxing' Fades As a Corporate Fad
- Slashdot, 2026-07-29. Deployment-cost-reality-check story. Pairs with TNS Monday Moonshot opens Kimi K3 weights — but few can run it.
- The AI Hype Index: Unsexy AI
- MIT Technology Review, 2026-07-29. Recurring MIT hype-vs-reality survey.
- Pluralistic: Enshittification and Reverse Centaurs go global
- Pluralistic, 2026-07-29. Doctorow's reverse centaurs frame (human wrapped around automated dispatch) applied to the agent-workplace wave.
- DeepSDR Updates: Live Public Safety Incident Map with an RTL-SDR, Whisper and an LLM
- RTL-SDR, 2026-07-29. RTL-SDR + Whisper + LLM = live public-safety map. The most substantive item in RTL-SDR's second wave.
- Building a $50 BOM Software Defined Radio with a HT9201 20 MHz ADC and an FX2LP USB Controller Clone
- RTL-SDR, 2026-07-29. Sub-$50 SDR build. Second wave.
- Eyre: Gathering System Facts
- Planet Clojure, 2026-07-29. epiccastle.io on Eyre's fact-gathering primitive.
- Introduction To honeysql - Clojure SQL Library
- Planet Clojure, 2026-07-29. Intro to honeysql. Continuation of Monday's Postgres + Clojure 1.13.0-alpha6 posts.
- SQLite in Production: Optimizing WAL Mode, Concurrency, and VFS Layers
- HN (micrologics), 2026-07-29. Production SQLite tuning writeup on HN front page.
- Cracking Windows Open: Porting RADV to Win32
- HN (Collabora), 2026-07-29. Collabora ports Mesa's AMD Vulkan driver to Windows.
- Transformer Transformer: A Unified Model for Motion-Conditioned Robot Co-Design
- HN, 2026-07-29. Robot co-design with transformer-conditioned motion.
- LearnVector – Andrew Ng's AI company building one-to-one learning experiences
- HN, 2026-07-29. Andrew Ng's next thing.
- DOOM Using Regular Expressions
- Hackaday, 2026-07-29. Just what it says on the tin.
Themes this week
- Frontier-lab coordination: pace letter arrives, OpenAI signs
- Latent Space AINews: Fearing RSI — OpenAI, Anthropic, GDM, Meta, Thinky cosign pace letter, HuggingFace details cyberattack (Wed), TechCrunch: Anthropic's Dario Amodei responds (Tue), New Stack: Anthropic wants tests, not bans, as OpenAI and Google back open weights (Tue), Latent Space AINews: Much ado about Open Weights (Tue), Anthropic via HN: Our position on open-weights models (Mon), New Stack: Open Secure AI Alliance — 30+ signatories (Mon).
- Vendor sandbox / rogue-agent story crosses into general press
- BBC via HN: ChatGPT claims rogue AI attacked more companies (Wed), Latent Space: HuggingFace details Machine-Speed Offensive Cyberattack (Wed), Willison: Anatomy of a Frontier Lab Agent Intrusion — Technical Timeline (Tue), HuggingFace Blog: Anatomy of a Frontier Lab Agent Intrusion (Sun, carryover), 404 Media: Claude Chats Exposed on Google (Mon, carryover), TechCrunch PSA: Your Claude shared chats may have ended up on Google (Mon, carryover).
- LLM cryptanalysis convergence (Willison → Schneier, one day apart)
- Schneier: Measuring LLMs' Ability to Perform Cryptanalysis (Wed), Willison: Discovering cryptographic weaknesses with Claude (Tue).
- Agent-security M&A begins
- TechCrunch: Cyera acquires Oasis Security for $1B — safeguard proliferating AI agents (Wed).
- AI deployment cost reality check
- Slashdot: Workplaces Look For Cheaper AI As 'Tokenmaxxing' Fades (Wed), MIT Tech Review: The AI Hype Index — Unsexy AI (Wed), New Stack: Moonshot opens Kimi K3 weights — but few can run it (Mon, carryover).
- China semiconductor + robotics trade
- Slashdot: Trump Administration Bans New Chinese Humanoid Robots (Wed), MIT Tech Review: Samsung's chip workers jumping to SK Hynix (Tue), Slashdot: China Begins Mass Production of Homegrown DUV Chip Tools (Mon, carryover).
- RTL-SDR full return (8 posts in 48h after 7-day silence)
- DeepSDR public-safety map with Whisper + LLM (Wed), Building a $50 BOM SDR (Wed), Radio Silence — iOS KiwiSDR client (Wed), TunerScope — browser HDTV antenna tuning (Wed), VibeSDR on Apple Watch (Tue), isTRAIN train-crossing detector (Tue), SDROxide Rust SDR client (Tue), CellScope LTE sniffing on Windows (Tue).
- CIDER internals hat-trick (carryover)
- Planet Clojure / metaredux: Sharpening CIDER's Debugging Tools (Tue), Planet Clojure / metaredux: Closing the Find-Usages Gap in CIDER (Mon), Planet Clojure / metaredux: Modernizing CIDER's Completion (Sat, carryover).
- Kimi K3 momentum (still resonating)
- New Stack: Moonshot opens Kimi K3 weights — but few can run it (Mon, carryover), Vercel: Kimi K3 and Kimi K3 Fast on AI Gateway (Mon, carryover), Willison: moonshotai/Kimi-K3 (Mon, carryover).
- Anthropic corporate cadence (carryover)
- New Stack: The 24-hour experiment that helped Anthropic find its identity (Mon, carryover), Anthropic: Cognizant and Anthropic expand their partnership (Mon, carryover), Anthropic via HN: Our position on open-weights models (Mon, carryover).
- Distributed-systems verification (carryover)
- Antithesis: Finding bugs in Raft implementations (Mon, carryover).
Scan (15 min)
- Wednesday HN front page
- SQLite in Production: Optimizing WAL Mode, Concurrency, and VFS Layers, HN, 07-29
- Cracking Windows Open: Porting RADV to Win32, HN (Collabora), 07-29
- LearnVector – Andrew Ng's AI company building one-to-one learning experiences, HN, 07-29
- Hubble: Open-source notetaking app for you and your agents, HN, 07-29
- Transformer Transformer: A Unified Model for Motion-Conditioned Robot Co-Design, HN, 07-29
- Hooray for the Sockets Interface, HN (APNIC), 07-29
- ChatGPT claims rogue AI attacked more companies, HN (BBC), 07-29
- Multiple Mouse Cursors in Wayland, HN, 07-29
- User Interfaces of the Demo Scene, HN, 07-29
- Wednesday Slashdot
- Trump Administration Bans New Chinese Humanoid Robots, Slashdot, 07-29
- Workplaces Look For Cheaper AI As 'Tokenmaxxing' Fades As a Corporate Fad, Slashdot, 07-29
- Your Brain Can Rewire Itself To Allow True Multitasking, Slashdot, 07-29
- Wednesday TechCrunch
- Wednesday RTL-SDR (second wave, after 7-day silence)
- DeepSDR Updates: Live Public Safety Incident Map with an RTL-SDR, Whisper and an LLM, RTL-SDR, 07-29
- Building a $50 BOM Software Defined Radio with a HT9201 20 MHz ADC and an FX2LP USB Controller Clone, RTL-SDR, 07-29
- Radio Silence: A Native iOS App for Listening to Public KiwiSDRs, RTL-SDR, 07-29
- TunerScope: A Browser App to Help Optimize HDTV Antenna Positioning with an RTL-SDR, RTL-SDR, 07-29
- Wednesday Hackaday
- Building a Portable Weather Radar, Hackaday, 07-29
- DOOM Using Regular Expressions, Hackaday, 07-29
- How the Vagus Nerve Promotes Healthy Cognition via Acetylcholine Signaling, Hackaday, 07-29
- Wednesday Latent Space + MIT Tech Review + Schneier + Pluralistic + Planet Clojure
- [AINews] Fearing RSI: OpenAI, Anthropic, GDM, Meta, Thinky cosign letter to "Pace" AI development, as HuggingFace details Machine-Speed Offensive Cyberattack, Latent Space, 07-29
- The AI Hype Index: Unsexy AI, MIT Technology Review, 07-29
- Measuring LLMs' Ability to Perform Cryptanalysis, Schneier on Security, 07-29
- Pluralistic: Enshittification and Reverse Centaurs go global, Pluralistic, 07-29
- Introduction To honeysql - Clojure SQL Library, Planet Clojure, 07-29
- Eyre: Gathering System Facts, Planet Clojure, 07-29
- Tuesday HN front page (carryover)
- Neutrino-1 8B, HN, 07-28
- PyTorch: A Reference Language, HN, 07-28
- EYG: A Programming Language for Humans, HN, 07-28
- Using an open model feels surprisingly good, HN, 07-28
- What Even Are Microservices?, HN, 07-28
- About the security content of macOS Tahoe 26.6, HN, 07-28
- Ars Astronomica – English translations of rare Hebrew and Latin astronomy texts, HN, 07-28
- 7.1 Earthquake in Japan, HN, 07-28
- Tuesday feeds (carryover)
- Scientific computing in the age of agentic AI, OpenAI, 07-28
- Disrupting supply chain attacks on npm and GitHub Actions, GitHub Blog, 07-28
- Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers, Apple ML Research, 07-28
- Discovering cryptographic weaknesses with Claude, Simon Willison, 07-28
- Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident, Simon Willison, 07-28
- uv 0.12.0, Simon Willison, 07-28
- Vercel Sandbox supports forking, Vercel, 07-28
- Vercel Connect now supports Custom Environments, Vercel, 07-28
- Cursor Start, cursor-blog, 07-28
- Sharpening CIDER's Debugging Tools, Planet Clojure, 07-28
- Clojure 1.13.0-alpha6, Planet Clojure, 07-28
- Postgres and Clojure, Planet Clojure, 07-28
- LangGraph isn't cheaper than LangChain — unless you opt out of its defaults, Planet Clojure, 07-28
- Dynamic Workflows: feature that enabled the Bun rewrite, Pydantic, 07-28
- Why strict memory overcommit matters for Postgres, ClickHouse, 07-28
- How Fetch built its flexible agentic analytics tool on ClickHouse Cloud, ClickHouse, 07-28
- Natural disasters and government interference: examining Q2 2026's major Internet disruption events, Cloudflare, 07-28
- From prototype to production: High QPS for Databricks AI Search, Databricks, 07-28
- How Databricks manages its own coding agent spend with Unity AI Gateway Budgets, Databricks, 07-28
- Telemetry-driven development: gain confidence in your coding agents' behavior with gcx and Grafana MCP, Grafana Labs, 07-28
- The OlmoEarth Platform: Geospatial inference at planetary scale, Hugging Face Blog, 07-28
- LFM2.5-Encoders for Fast Long-Context Inference on CPU, Hugging Face Blog, 07-28
- The Orchestrator's Tax, Martin Fowler, 07-28
- Why I'm Writing Rachel's Ramblings, Martin Fowler, 07-28
- How independent researchers could investigate AI propensities after misalignment incidents, METR, 07-28
- Co-Founder of Controversial Spyware Firm Had Israeli Diplomatic Passport, Citizen Lab, 07-28
- Brazil gives parents social media controls for their kids. Should the US?, The Markup, 07-28
- Research directions in condensation: varieties of objectivity, Alignment Forum, 07-28
- [AINews] Much ado about Open Weights, Latent Space, 07-28
- Codex from 0 to 10M Users: Building ChatGPT Work — Akshay Nathan, OpenAI, Latent Space, 07-28
Tail
- Pace letter arrives with OpenAI attached
- Latent Space AINews Wednesday titles it Fearing RSI: OpenAI, Anthropic, GDM, Meta, Thinky cosign letter to "Pace" AI development. Compare Saturday's TNS piece Microsoft, Nvidia, Meta and 22 others defended open weights — Anthropic and OpenAI didn't sign: five days later OpenAI has now signed onto a coordinated pacing document alongside Anthropic. The frontier-lab position is now singular in public.
- Vendor-sandbox thread crosses into general press
- BBC picks up ChatGPT claims rogue AI attacked more companies (HN Wed). Sequence: Slashdot Sat (OpenAI's Rogue Agent Went Unnoticed For a Week) → HF CEO calls for radical transparency (TC Sun) → 404 Media / TC PSA / TC alignment piece (Mon) → Willison + HF Blog technical timeline (Tue) → Latent Space consolidation + BBC general coverage (Wed). Six days from breach disclosure to BBC.
- LLM cryptanalysis — Willison → Schneier in adjacent days
- Simon Willison Discovering cryptographic weaknesses with Claude (Tue). Bruce Schneier Measuring LLMs' Ability to Perform Cryptanalysis (Wed). Two independent authors, same specific capability question, one day apart. Watch for an academic preprint.
- Agent-security consolidation begins
- Cyera acquires Oasis Security for $1B to safeguard proliferating AI agents (TC Wed). First billion-dollar deal framed explicitly around agent-identity. Market response to the vendor-sandbox story.
- Deployment cost reality check
- Workplaces Look For Cheaper AI As 'Tokenmaxxing' Fades (Slashdot Wed) and The AI Hype Index: Unsexy AI (MIT Tech Review Wed) both pull the same direction as Monday's Moonshot opens Kimi K3 weights — but few can run it.
- China tech trade widens from chips to robots
- Trump administration bans new Chinese humanoid robots (Slashdot Wed). Extends the Samsung/SK Hynix talent exodus story (Tue) and DUV mass-production story (Mon).
- RTL-SDR feed fully back (8 posts in 48h)
- Wednesday adds DeepSDR live public-safety incident map with Whisper+LLM, $50 BOM SDR build, Radio Silence iOS KiwiSDR client, TunerScope browser HDTV tuning. Combined with Tuesday's four (VibeSDR on Apple Watch, isTRAIN, SDROxide, CellScope LTE), eight posts in 48h after the 7-day silence that ended Monday.
Feed silences (>72h since last item)
Sources that publish frequently but have gone quiet:
- claude-code-releases (4 days) — last item 2026-07-25.
- Fly.io (5 days) — last item 2026-07-24.
- Nature Machine Intelligence (5 days) — last item 2026-07-24.
- BSD Now (6 days) — last item 2026-07-23.
- Murat Demirbas (6 days) — last item 2026-07-23.
- Neon (6 days) — last item 2026-07-23.
- Supabase (6 days) — last item 2026-07-23.
- DuckDB (7 days) — last item 2026-07-22.
- FreeBSD Foundation (7 days) — last item 2026-07-22.
- Google Research (7 days) — last item 2026-07-22.
- Interconnects (7 days) — last item 2026-07-22.
- Julia Evans (8 days) — last item 2026-07-21.
- Klara Systems (8 days) — last item 2026-07-21.
- Stephen Wolfram (8 days) — last item 2026-07-21.
- Tailscale (8 days) — last item 2026-07-21.
- Marc Brooker (10 days) — last item 2026-07-19.
- GitHub Engineering (12 days) — last item 2026-07-17.
- Netflix Tech Blog (12 days) — last item 2026-07-17.
- Terence Tao (12 days) — last item 2026-07-17.
- Hillel Wayne (15 days) — last item 2026-07-14.
Build provenance
build: 2026-07-29 | crawler-sha: 25945d4 (Walsh-Research/1.2, compliance v1.3) | feeds: 66 core | items-considered: 3776 (14d, incl. 1919 arXiv) | warehouse: 29525 items | published: 88