Morning Brief: Friday, July 10

Sixty-five feeds. Two weeks. 4,208 items reduced to what follows. (what we track, how we crawl, subscribe)

Friday. Latent Space packages the OpenAI day: GPT 5.6 Sol/Terra/Luna ships, Codex becomes the ChatGPT superapp. TechCrunch runs the family announcement; Slashdot frames the release as post-government greenlight and confirms the ChatGPT Work launch; a companion TechCrunch piece asks how the government decided the frontier model was safe. Also this morning: Fidji Simo steps down from OpenAI's no. 2 role; Atlas is shut down but the browser ambitions continue; NYT accuses OpenAI of hiding evidence in the ChatGPT copyright trial; GPT-5.6 is declared the preferred model in Microsoft 365 Copilot amid breakup chatter. Meta answers with Muse Spark 1.1, its first paid model, and enters the AI-coding arena. Slashdot flags a Nobel-winning US chemist relocating to China to lead an AI institute, and lawmakers probing Chinese AI model use in US companies. Alignment Forum publishes on value correction and NL-autoencoder robustness. EPFL's Nevo project generates videos tuned to drive a specific target brain region. Databricks names the ambulatory intelligence gap. On the tail: Pluralistic on rights for robots, Emacs-as-service, and mechanosynthesis of atomic carbon via inverted-mode STM.

Top (5-7 min)

OpenAI launches GPT 5.6 Sol/Terra/Luna, Codex becomes ChatGPT superapp: Latent Space, 2026-07-10. AINews packages the OpenAI day: a three-variant GPT-5.6 family and the folding of Codex into ChatGPT as the Work superapp.
OpenAI launches its new family of models with GPT-5.6: TechCrunch, 2026-07-10. The straight release note: GPT-5.6 ships as a family, with Sol, Terra and Luna as the tier names.
OpenAI rolls out GPT-5.6 after government greenlight, announces ChatGPT Work: Slashdot, 2026-07-10. Slashdot frames the release around the government greenlight and the parallel ChatGPT Work launch — the Codex-superapp face for enterprise.
How did the government decide OpenAI's frontier model was safe to release?: TechCrunch, 2026-07-10. Companion piece: interrogates the safety-review process that produced the greenlight, and what it says about the evaluation pipeline.
Fidji Simo steps down from OpenAI's no. 2 role: TechCrunch, 2026-07-10. A leadership exit on the same day as the model family launch — worth reading against the Codex-into-ChatGPT reorg.
OpenAI is shutting down Atlas, but its AI browser ambitions are still growing: TechCrunch, 2026-07-10. Product consolidation: Atlas closes, browser ambitions continue — probably absorbed into the superapp surface.
New York Times says OpenAI hid evidence in ChatGPT copyright trial: TechCrunch, 2026-07-10. Legal pressure lands the same morning as the launch: NYT accuses OpenAI of evidence suppression in the copyright suit.
Meta enters the crowded AI coding battle with Muse Spark 1.1: TechCrunch, 2026-07-10. Meta's first paid AI model, aimed at the coding market — sets up Muse-vs-Codex-vs-Claude-Cowork as the enterprise coding triple.

Themes this week

OpenAI's big day and the superapp pattern: Latent Space: GPT-5.6 Sol/Terra/Luna, Codex becomes ChatGPT superapp (today), TechCrunch: new family of models with GPT-5.6 (today), Slashdot: GPT-5.6 after government greenlight, ChatGPT Work (today), Willison: quoting OpenAI on Work-cloud-vs-desktop scoping (today), TechCrunch: Atlas is shut down (today), Latent Space: Grok 4.5, first Opus-class model post-Cursor (Thu).
OpenAI leadership and legal pressure: Simo steps down from OpenAI no. 2 (today), NYT accuses OpenAI of hiding evidence (today), TC: how did government decide GPT-5.6 was safe (today), TC: GPT-5.6 preferred in MS 365 Copilot amid breakup chatter (today).
Meta enters the coding wars: TC: Muse Spark 1.1, Meta's first paid AI model (today), TNS: Meta debuts Muse Spark 1.1 (Thu), Willison: Introducing Muse Spark 1.1 (Thu), TC: Meta's new AI chips begin production in September (today).
US-China AI decoupling: Slashdot: Nobel-winning US chemist to China to lead AI institute (today), Slashdot: lawmakers probe Chinese AI models in US companies (today).
Alignment theory and interpretability: Alignment Forum: value generalisation — value correction (today), Alignment Forum: how robust are natural-language autoencoders to initialization (today), Alignment Forum: $160M grant from Coefficient Giving (Thu).
Neuro-AI: driving brains and monitoring parents: EPFL Nevo: AI-generated videos to maximally drive a target brain region (today), Frontiers: parental device use and adolescent-caregiver attachment bond (today), MIT MTL: graphene thin-film for neural interfaces (today).
AI in biology and healthcare: Databricks: the ambulatory intelligence gap (today), Nature MI: four questions for AI-ready biological data (today), MIT Tech Review: European fertility group calls for donor limits (today).

Scan (15 min)

Friday signal (short, worth reading)
- OpenAI launches GPT 5.6 Sol/Terra/Luna, Codex becomes ChatGPT superapp, Latent Space, 07-10
- OpenAI rolls out GPT-5.6 after government greenlight, announces ChatGPT Work, Slashdot, 07-10
- OpenAI launches its new family of models with GPT-5.6, TechCrunch, 07-10
- How did the government decide OpenAI's frontier model was safe to release?, TechCrunch, 07-10
- Fidji Simo steps down from OpenAI's no. 2 role, TechCrunch, 07-10
- Quoting OpenAI, Simon Willison, 07-10
OpenAI ripple effects
- OpenAI is shutting down Atlas, but its AI browser ambitions are still growing, TechCrunch, 07-10
- NYT says OpenAI hid evidence in ChatGPT copyright trial, TechCrunch, 07-10
- GPT-5.6 the 'preferred model' for MS 365 Copilot amid breakup chatter, TechCrunch, 07-10
- How Deutsche Telekom is rewiring telecommunications with AI, OpenAI, 07-10
Meta and the coding-agent market
- Meta enters the crowded AI coding battle with Muse Spark 1.1, TechCrunch, 07-10
- Meta's new AI chips will begin production in September, TechCrunch, 07-10
- Instagram users: how to stop Meta's AI from using your photos, TechCrunch, 07-10
Alignment, safety, evaluation
- Value generalisation: value correction, Alignment Forum, 07-10
- How robust are natural language autoencoders to initialization?, Alignment Forum, 07-10
US-China and AI industrial policy
- Nobel-Winning US Chemist Will Move to China to Lead AI Institute, Slashdot, 07-10
- Lawmakers Probe Growing Use of Chinese AI Models In US Companies, Slashdot, 07-10
- An AI agent startup just let its agent run its $100M fundraise, TechCrunch, 07-10
- Mercor is in talks for a $20B valuation, TechCrunch, 07-10
Neuroscience, biology, healthcare
- AI-generated videos to maximally drive a target brain region, HN, 07-10
- Parental device use and the adolescent-caregiver attachment bond, HN, 07-10
- Graphene thin film technology for neural interfaces, MIT calendar, 07-10
- The Ambulatory Intelligence Gap, Databricks, 07-10
- Four questions for AI-ready biological data, Nature MI, 07-10
- Sperm donors need limits, says a European fertility group, MIT Tech Review, 07-10
Robotics, surgery, humanoids
- Humanoid Robots Controlled By Surgeons Did World-First Operation On Live Pigs, Slashdot, 07-10
Tooling, systems, ML infra
- Profiling in PyTorch (Part 3): Attention is all you profile, HF Blog, 07-10
- In Emacs, Everything Looks Like a Service, HN, 07-10
Hackaday build log
- Mechanosynthesis of Atomic Carbon Structures Using Inverted-Mode STM, Hackaday, 07-10
- A Super Cheap Desk Toy Becomes a Hackable Desktop Notifier, Hackaday, 07-10
- Three Different Digital Counters To Remind Us How Good We Have It, Hackaday, 07-10
Aviation, media, culture
- Bjorn's Corner: Aircraft Structures Part 9. The Composite Matrix., Leeham, 07-10
- Pluralistic: "Rights for robots" and the AI slavery fantasy, Pluralistic, 07-10
Yesterday's carryover worth re-reading Friday morning
- SpaceXAI launches Grok 4.5, first Opus-class model post-Cursor, Latent Space, 07-09
- Postgres rewritten in Rust, passing 100% of the regression tests, HN, 07-09
- Google's new remote attestation scheme is every bit as terrible as its old, EFF, 07-09
- $160M grant from Coefficient Giving, Alignment Forum, 07-09
- Searchable field-level encryption on Supabase with CipherStash, Supabase, 07-09

Tail

Claude Code v2.1.206, claude-code-releases, 07-10
How GitHub gave every repository a durable owner, Pinboard, 07-10
Central Campus Public Art Tour, MIT calendar, 07-10
Outdoor Movie: Fantastic Mr. Fox, MIT calendar, 07-10

Feed silences (diagnostic)

arxiv-cs-ai: 0 items today. Friday cliff: after Wed 234, Thu 202, the weekend/Friday drop-off has arrived. Expected — arXiv typically thins into the weekend, and Monday will absorb the delta.
Anthropic first-party, OpenAI (blog vs Slashdot/TechCrunch coverage), Simon Willison (aside from the OpenAI quote), Apple ML Research, Google Research, ClickHouse, EFF Deeplinks, The New Stack, LWN, Planet Clojure, Schneier, GitHub Blog, Interconnects, Kenneth Payne, Martin Fowler, Microsoft Research, Cloudflare, Charity Majors: no primary posts today.
Netflix Tech Blog, James Bornholt: persistent DNS/TLS errors continue (unchanged).
WBUR: no Endless Thread today.
bitsavers (6 feeds): connected, 0 items (sparse archive, expected).
claude-code-releases: v2.1.206 shipped today.

Build provenance

build: 2026-07-10 | crawler-sha: 5fe7ab8 (Walsh-Research/1.2, compliance v1.3) | feeds: 65 active (78 configured, incl. 12 corp-eng, 6 bitsavers, 5 generated) | items-considered: 4208 (14d, incl. 2432 arXiv) | warehouse: 23863 items | published: 8 | note: Friday. OpenAI's big day — GPT-5.6 in three variants (Sol/Terra/Luna), Codex folds into ChatGPT as Work, government greenlight, Atlas shutdown, Fidji Simo steps down, NYT accuses OpenAI of hiding evidence, GPT-5.6 preferred in MS 365 Copilot. Meta answers with Muse Spark 1.1. Nobel-winning US chemist moves to China for an AI institute. Alignment Forum: value correction, NL-autoencoder robustness. EPFL Nevo generates videos to maximally drive a target brain region. Databricks names the ambulatory intelligence gap. arxiv-cs-ai: 0 items (Friday cliff).