Morning Brief: Monday, June 29
Sixty-six feeds. Two weeks. 4,993 items reduced to what follows. (what we track, how we crawl, subscribe)
Monday. arXiv returns: ~100 new cs.AI papers, of which the agentic-systems cluster is the densest in weeks — repository-level governance, agent-native immune systems, tandem RL with verifiable rewards, and (notably) Google's own paper-assistant tool for automating scientific review. The last item lands on the same morning Ars Technica covers two 1940s Max Planck papers retracted by what looks like a rogue retraction algorithm. The institutional substrate of peer review and the AI substrate of automated review are entering the same room.
Underneath it: SemiAnalysis estimates 40GW+ of behind-the-meter US datacenter capacity by 2028, with the grid as the binding constraint rather than the chips. Pairs directly with Sunday's token-cost-versus-salary curve and Friday's Codex internal-usage 56x. The token-economics story has a wattage bill that is now visible.
Quieter: a Trump-shuttered NOAA climate change site returns under nonprofit stewardship — the first cleanly successful "civil society absorbs a defunded federal data asset" pattern. Age verification keeps tightening: a piece argues the next state is automated attribution of speech, not gatekeeping. The Pragmatic Engineer reports Pollen tried to remove an article about its CEO and CTO, and Google assisted — Streisand 2.0.
Top (5-7 min)
- US Grid Constraints: Towards 40GW+ of Behind-the-Meter Datacenter by 2028?
- HN, 2026-06-29. SemiAnalysis. The physical-infrastructure substrate under the AI cost curve. Pairs with Sunday's developer token costs exceeding salaries and Friday's Codex internal 56x growth. The marginal token has a wattage bill.
- Why did this journal retract two 1940s papers by Max Planck?
- HN, 2026-06-29. Plus Slashdot: Max Planck slapped with two paper retractions by suspected rogue algorithm. Lands the same day arXiv surfaces Google's Paper Assistant Tool for automating scientific review. Automated review meets automated retraction.
- Trump-Shuttered Climate Change Site Now Back Online In Nonprofit Hands
- Slashdot, 2026-06-29. First clean instance of "civil society absorbs a defunded federal data asset." A template for what the climate-data-preservation thread has been waiting for.
- Age verification is just a precursor to automated attribution of speech
- HN, 2026-06-29. Reframes the age-verification fight as the first beachhead for tying speech to identity at the infrastructure level. Adds to Schneier and EFF's running thread on surveillance creep.
- Pollen tried to remove my article about CEO Negus-Fancey, and Google is assisting
- HN, 2026-06-29. Pragmatic Engineer. Streisand 2.0 — failed-fintech press laundering attempt with Google in the loop. Pairs with Sunday's Pluralistic on Zuckerberg vs whistleblowers.
- Govern the Repository, Not the Agent: Measuring Ecosystem-Level Risk in AI-Native Software
- arXiv, 2026-06-29. Operationalizes the thesis the Akrites coalition is sitting on. Repository-level risk metrics rather than per-agent guardrails. Direct successor to last week's Akrites OSS-security launch.
Themes this week
- Token cost meets grid cost
- SemiAnalysis: 40GW+ behind-the-meter datacenter by 2028 (today), Developer token costs may exceed salaries in two years (Sun), OpenAI internal Codex 56x growth (Fri), Renewable energy hits 30% of US electricity generation (Sun), TNS: session as the new unit of compute (Fri). Three layers of the same cost stack get visible in one week.
- Restricted release becomes bilateral
- GPT-5.6 Sol/Terra/Luna restricted to trusted partners (Fri), METR predeployment evaluation of Sol (Fri), USG authorizes Anthropic Mythos for trusted orgs (Sat), TechCrunch: ~100 companies and agencies on the Mythos allow-list (Sat), OpenAI Deployment Safety Hub for GPT-5.6 Preview (Sun), Asian AI startups launch Mythos-like models (Sat), China's AI matches Anthropic in cybersecurity, raising US-restriction worries (Sun).
- Automated review meets automated retraction
- arXiv: Towards Automating Scientific Review with Google's Paper Assistant Tool (today), Ars Technica: Why two 1940s Max Planck papers were retracted (today), Slashdot: suspected rogue retraction algorithm (Sat), El País: mass AI fraud on Brown University exam (Sun).
- Surveillance creep, attribution, and removal
- Age verification → automated attribution of speech (today), Pollen + Google try to remove Pragmatic Engineer article (today), Pluralistic: Zuckerberg vs whistleblowers (Sat), Schneier: Meta tests facial recognition for police and military (Fri), US agency cancels warrantless mobile-tracking contract (Sun).
- Repository-as-unit, agent governance papers land
- Govern the Repository, Not the Agent (arXiv) (today), Agent-Native Immune System: Architecture, Taxonomy, and Engineering (arXiv) (today), Agentic Hardware Design as Repository-Level Code Evolution (arXiv) (today), Tandem RL with Verifiable Rewards (arXiv) (today), Akrites vulnerability-mitigation project launches (LWN) (Fri).
Scan (15 min)
- Grid, energy, datacenter physics
- US Grid Constraints: Towards 40GW+ Behind-the-Meter Datacenter by 2028, SemiAnalysis via HN, 06-29
- Renewable Energy Just Hit 30% of America's Electricity Generation, Slashdot, 06-28
- SpaceX plans 'Starpipe' natural gas pipeline for Starship, Slashdot, 06-27
- Agents, governance, harness papers (arXiv Monday)
- Govern the Repository, Not the Agent: Ecosystem-Level Risk in AI-Native Software, arXiv, 06-29
- Agent-Native Immune System: Architecture, Taxonomy, and Engineering, arXiv, 06-29
- Agentic Hardware Design as Repository-Level Code Evolution, arXiv, 06-29
- Tandem Reinforcement Learning with Verifiable Rewards, arXiv, 06-29
- Internalizing the Future: A Unified Agentic Training Paradigm for World Model Planning, arXiv, 06-29
- ATOD: Annealed Turn-aware On-policy Distillation for Multi-turn Autonomous Agents, arXiv, 06-29
- NormAct: Hidden Social Norm Compliance in Embodied Planning, arXiv, 06-29
- Grounded Iterative Language Planning: Parameterized World Models Reduce Hallucination Propagation, arXiv, 06-29
- Cost, context, slop
- Developer AI Token Costs Could Exceed Their Salaries, Slashdot, 06-28
- Tokenmaxxing is dead, long live tokenmaxxing, HN, 06-28
- Vibe slop is the symptom. Context debt is the disease., TNS, 06-27
- Greptile, Cursor, Devin agree agents should run their code, TNS, 06-27
- Eval, safety, governance
- Semgrep: GLM 5.2 beats Claude in cyber benchmarks, HN, 06-28
- Deployment Awareness Matters More Than Evaluation Awareness, Alignment Forum, 06-26
- The Case for Model Forensics, Alignment Forum, 06-26
- Harmless image can jailbreak vision-language models, Slashdot, 06-27
- Post-Mythos Cybersecurity: Keep calm and carry on, HN, 06-27
- Scientific publishing and AI-driven review
- Towards Automating Scientific Review with Google's Paper Assistant Tool, arXiv, 06-29
- Why did this journal retract two 1940s papers by Max Planck?, Ars Technica via HN, 06-29
- Max Planck Slapped With Two Paper Retractions By Suspected Rogue Algorithm, Slashdot, 06-27
- Professor denounces mass AI fraud on an exam at Brown, El País via HN, 06-28
- Surveillance, attribution, and platform power
- Age verification → automated attribution of speech, nonogra.ph via HN, 06-29
- Pollen + Google try to remove Pragmatic Engineer article, HN, 06-29
- Meta Is Testing Facial Recognition for Police and Military, Schneier, 06-26
- One Million Passports Leaked Online, Schneier, 06-26
- Climate, civil society absorption
- Trump-Shuttered Climate Change Site Back Online In Nonprofit Hands, Slashdot, 06-29
- France's Heat This Week Was Worse Than a Dire Scenario Imagined For 2050, Slashdot, 06-27
- Heat waves mess with your brain, MIT Tech Review, 06-26
- Tools and infrastructure
- Aviation, space, science
- Airbus's 'China gap' for the A220, Leeham, 06-29
- Open Forum, Week of June 29, Leeham, 06-29
- Scientists Think Neptune and Uranus May Not Be the Ice Giants We Imagined, Slashdot, 06-29
- First U.S. deployment of a collaborative combat aircraft, The Air Current, 06-26
- Languages, runtime, kernel
- Kernel prepatch 7.2-rc1, LWN, 06-28
- Clojure on Fennel part four: Parsing (again), Planet Clojure, 06-28
- Latest open artifacts (#22): Zyphra, Cohere, Poolside, Interconnects, 06-28
- Hack Your Summer, Simon Willison, 06-28
Tail
- Daisugi, the Japanese technique of growing trees out of other trees, HN, 06-28
- The Baffling World of Masayoshi Son's Presentations, HN, 06-26
- Historical memory prices 1960-2026, HN, 06-28
- 5k menus from the New York Public Library's Buttolph Collection, HN, 06-28
- Pre-Modern Armies for Worldbuilders, Part III: Paying for It, HN, 06-26
Feed silences (diagnostic)
arxiv-cs-ai: returned today, ~100 items.Anthropicfirst-party RSS: still 404; covered viaanthropic-generated+ Latent Space + TNS + Slashdot/HN aggregation.bitsavers(6 feeds): connected, 0 items (sparse archive).James Bornholt,Netflix Tech Blog: persistent DNS/TLS errors.claude-auth: brief generation auth failed every morning June 21–28; manual regeneration continues today.
Build provenance
build: 2026-06-29 | crawler-sha: 5fe7ab8 (Walsh-Research/1.2, compliance v1.3) | feeds: 66 active (78 configured, incl. 12 corp-eng, 6 bitsavers, 5 generated) | items-considered: 4993 (14d, incl. 3132 arXiv) | warehouse: 19993 items | published: 52 | note: arXiv returns Monday with dense agentic-governance cluster (Govern the Repository / Agent-Native Immune System / Tandem RL with Verifiable Rewards); SemiAnalysis 40GW behind-the-meter datacenter forecast threads under Sunday's token-cost-vs-salary story; Trump-shuttered NOAA climate site returns under nonprofit stewardship; Max Planck retractions land same day as arXiv Google-paper-assistant tool