Key Highlights
- Anthropic’s $1.5B copyright settlement wins final court approval — the largest in U.S. history, paying roughly $3,000 per work across ~500,000 titles. The ruling upheld AI training as fair use but penalized how Anthropic sourced books (pirate sites like Library Genesis).
- A Nikkei investigation pegs five U.S. tech giants’ hidden, off-balance-sheet AI debts at ~$1.65 trillion — Meta’s alone reaches ~$420B, nearly triple its reported liabilities. Data-center leases and GPU supply deals are being structured outside conventional disclosures, clouding real leverage.
- Claude Fable 5 reportedly produced a one-line formula resolving the 87-year-old Jacobian conjecture, one of several long-standing math problems AI models have cracked this year.
- Boris (Claude Code co-creator) argues that encoding domain knowledge as infrastructure — CLAUDE.md files, skills, lint rules, tests — is now the highest-leverage engineering skill, unpacked in a t3dotgg video on why senior-level impact comes from building systems others (and their agents) can contribute through.
- Open-weight models are becoming a U.S. policy flashpoint — Moonshot’s Kimi K3 reignited debate over whether to discourage or even ban advanced Chinese models, pitting frontier-lab margins against open innovation.
Analysis & Opinion
Anthropic’s landmark $1.5B copyright settlement is approved — TechCrunch
A federal judge gave final approval to Anthropic’s $1.5 billion settlement with authors and publishers, clearing the way for payments of roughly $3,000 per work across an estimated 500,000 copyrighted titles — the largest copyright settlement in U.S. history. The court held that training AI on copyrighted text is fair use, but faulted Anthropic for sourcing books partly through pirate sites including Library Genesis rather than only legitimate purchases and scans. Judge William Alsup had granted preliminary approval before retiring; successor Judge Araceli Martinez-Olguin signed off. Many creators remain dissatisfied, noting the resolution addressed how books were obtained rather than establishing broader precedent on copyright and AI training.
Five US tech giants’ hidden debts soar to $1.65T on opaque AI funding — Nikkei (via Hacker News)
A Nikkei investigation finds undisclosed liabilities at five major U.S. tech companies have ballooned to roughly $1.65 trillion over four years, driven largely by AI infrastructure — data-center leases and GPU supply agreements structured off the balance sheet. Meta’s concealed debt alone is estimated at ~$420 billion, nearly triple its transparent liabilities. These obligations now dwarf the firms’ traditional reported debt, complicating investor risk assessment. The findings raise pointed questions about true leverage and vulnerability across the AI sector as capital deployment accelerates faster than disclosure norms.
OpenAI is scared of open-weight models. Should the US be? — TechCrunch
Chinese lab Moonshot’s advanced open-weight Kimi K3 has reignited debate over American AI companies’ economic interests versus broader technological progress. OpenAI’s Dean W. Ball initially argued for regulatory measures to discourage open-weight models — claiming they would reduce frontier-lab capital spending — drawing backlash from researchers like Yann LeCun and Martin Casado, who countered that open software drives innovation. Ball later retracted the claims, but reports suggest the Trump administration may weigh banning advanced Chinese models at the request of U.S. frontier companies. The core tension: cheaper open-weight alternatives running on independent infrastructure compress margins for Anthropic and OpenAI, even as they likely expand overall AI adoption rather than shrink it.
Trump’s latest AI czar has already resigned — TechCrunch
Chris Fall, director of the Center for AI Standards and Innovation (CAISI), has resigned after just three months. He followed Collin Burns, who lasted less than a week and was reportedly pushed out over his prior employment at Anthropic amid tensions with the administration. Before them, venture capitalist David Sacks led AI and crypto policy before stepping down in March. CAISI, housed under NIST, is the primary body developing technical standards, testing methods, and cybersecurity-risk assessments for AI models — leaving a repeatedly vacated seat at the center of U.S. AI governance.
YouTube clarifies policies around AI slop and upsetting videos — TechCrunch
YouTube tightened its monetization guidelines to define which AI-generated and low-quality videos can’t earn ad revenue, expanding its earlier “inauthentic content” rules. The revised policy names three categories: generic/repetitive/template-based content; off-putting or distressing material; and AI personas discussing sensitive topics like health and finance. Trust-and-safety chief Matt Halprin framed AI-enabled mass production as generic content that “doesn’t really show your creativity,” with the crackdown aimed at content farms. The move reflects YouTube’s financial stakes — it now out-earns rival streamers on ad revenue and has overtaken Netflix in average daily views globally.
Anthropic’s Fable survives the subscription axe — Rundown
Anthropic will keep Claude Fable 5 on Max and Team Premium plans permanently, but at half the standard usage caps, ending a five-week stretch of postponed removal deadlines. Lower-tier subscribers get a one-time $100 credit before shifting to pay-per-use access, with Anthropic citing demand that was “challenging to predict.”
Stop Using /init for AGENTS.md — Addy Osmani (Medium)
Addy Osmani argues that auto-generating a single root AGENTS.md via /init can slow agents down and raise costs without improving accuracy — one ETH Zurich study found LLM-generated context files cut task success 2–3% while increasing cost over 20%, largely due to redundancy agents could discover on their own. The fix is a scoped hierarchy of AGENTS.md files at the module level rather than a monolithic file mixing concerns.
AI’s most important protocol is getting a little bit easier to use — TechCrunch
An upcoming Model Context Protocol (MCP) update, arriving next week, shifts session handling to a “stateless” model that mirrors how conventional websites work, dropping the need for servers to track session IDs across distributed networks. The change should simplify maintenance and cut operational costs for companies running MCP servers, even if end users won’t notice directly.
New Products & Tools
Google is working on a new AI chip designed to make Gemini more efficient — TechCrunch
Alphabet is reportedly designing a server chip codenamed “Frozen v2,” targeting 2028 and up to 6–10x better efficiency (tokens per watt) than current chips, as firms race to build proprietary silicon amid AI capacity constraints.
5 ways to build a side hustle with Gemini — Google
Google pitches Gemini as an end-to-end tool for launching a small business, using notebooks and “business notebooks” to consolidate plans, communications, and Google Business Profile data while proactively surfacing action items.
At SIGGRAPH, NVIDIA Advances Graphics and Simulation With Agentic and Physical AI — NVIDIA Blog
At SIGGRAPH 2026, NVIDIA showcased neural rendering, world models, and simulation for building physically accurate virtual worlds, including 3D-guided neural rendering that preserves artistic intent, ensures temporal stability, and renders 4K in real time.
NVIDIA Agent Toolkit Expands With New Omniverse Libraries — NVIDIA Newsroom
NVIDIA folded Omniverse libraries into its Agent Toolkit so AI agents can inspect scenes, fix issues, and finalize simulation-ready 3D assets; Jensen Huang says “the physical AI era will be built in simulation first.” SideFX and PTC are among early adopters.
Integrate NVIDIA Omniverse RTX Sensor Simulation Into Existing Apps — NVIDIA Developer
NVIDIA’s ovrtx library (prerelease on GitHub) offers a lightweight C/Python SDK for generating camera, lidar, and radar outputs from OpenUSD scenes, letting apps and agents add physical-AI sensor simulation without leaving their host workflow.
NVIDIA NVLink: The Scale-Up Network for AI Factories — NVIDIA Developer
NVIDIA positions NVLink as purpose-built scale-up interconnect for “AI factories,” arguing that trillion-parameter MoE models and long-context reasoning make high-bandwidth, low-latency GPU-to-GPU communication the decisive architectural choice.
Bristol Myers Squibb Building Life Science Industry’s Most Advanced AI Factory on NVIDIA Vera Rubin — NVIDIA Blog
BMS is deploying a second DGX SuperPOD (eight DGX Vera Rubin NVL72 systems) delivering up to 10x performance per megawatt, aiming to give every scientist access to agentic drug-discovery workflows via the BioNeMo Agent Toolkit.
Gritt exits stealth with $34 million for robots to build solar plants — TechCrunch
Gritt, founded by two Carnegie Mellon roboticists, raised a $26M Series A (led by Obvious Ventures; $34M total) to deploy off-the-shelf hardware controlled by its AI models for solar-plant construction, betting new AI models can finally handle unstructured environments.
X relaunches a rebuilt Android app after year-long effort — TechCrunch
X shipped a ground-up Android rebuild on a modern Kotlin/Jetpack Compose stack, promising faster loading, scrolling, and notifications; Nikita Bier called it one of the largest engineering projects in the company’s history.
Adobe camera app’s new feature will critique your photos using AI — TechCrunch
Adobe’s experimental Project Indigo iOS app now uses LLMs to critique photos on composition, lighting, color, and emotional resonance, plus offer capture/editing guidance — favoring predictable button controls over open-ended prompts.
Research
Claude disproves an 87-year-old math problem — Rundown
Anthropic researcher Levent Alpöge reported that Claude Fable 5 produced a one-line formula resolving the Jacobian conjecture — open since 1939 — with the answer succinct enough for mathematicians to verify directly. It follows a year of AI models cracking long-standing problems, including an 80-year-old unit-distance problem and several Erdős problems.
Agent swarms and the new model economics — Cursor
Cursor’s redesigned agent-swarm architecture outperformed its predecessor across model configurations on a hard task (implementing SQLite in Rust from docs only) — reaching 80% in four hours with Grok 4.5 where the old swarm stalled. Different model mixes produced comparable quality at substantially different costs, pointing toward role-specialized planning/execution setups.
Interviews & Conversations
Boris has some really good advice — t3dotgg (19 min)
In this transcript-based video, Theo unpacks a post by Boris (a co-creator of Claude Code) arguing that classic high-leverage engineering habits — automating your workflow, writing lint rules, building test suites — matter more in the agent era, because every improvement multiplies across an army of agents. The central thesis: teams should convert domain knowledge that “lives in people’s heads” into infrastructure — CLAUDE.md/AGENTS.md steering files, skills, comments, and tests — so agents (and new human contributors, even non-engineers) can be productive with minimal prompting. Theo adds a strong caveat mirrored from experience: don’t let agents write your CLAUDE.md/AGENTS.md themselves, and don’t bulk-install skills before you’ve seen what actually breaks — build guardrails reactively after observing failures. He frames these steering files not as maps of “where things are” but as systems that push agents toward success, and argues that building such environments is now the real path to senior/staff-level impact. Notably, he links this to Boris’s line that “every team should be writing the CLAUDE.mds, review MDs, skills and docs that enable agents to productively work in their codebase with zero additional context from the prompter.”
References
- Anthropic’s landmark $1.5B copyright settlement is approved — TechCrunch, 2026-07-20 [blog]
- Five US tech giants’ hidden debts soar to $1.65T on opaque AI funding — Nikkei (via Hacker News), 2026-07-21 [blog]
- OpenAI is scared of open-weight models. Should the US be? — TechCrunch, 2026-07-20 [blog]
- Trump’s latest AI czar has already resigned — TechCrunch, 2026-07-20 [blog]
- YouTube clarifies policies around AI slop and upsetting videos — TechCrunch, 2026-07-20 [blog]
- Anthropic’s Fable survives the subscription axe — Rundown, 2026-07-20 [blog]
- Stop Using /init for AGENTS.md — Addy Osmani (Medium), 2026-07-21 [blog]
- AI’s most important protocol is getting a little bit easier to use — TechCrunch, 2026-07-20 [blog]
- Google is working on a new AI chip designed to make Gemini more efficient — TechCrunch, 2026-07-20 [blog]
- 5 ways to build a side hustle with Gemini — Google, 2026-07-20 [blog]
- At SIGGRAPH, NVIDIA Advances Graphics and Simulation With Agentic and Physical AI — NVIDIA Blog, 2026-07-20 [blog]
- NVIDIA Agent Toolkit Expands With New Omniverse Libraries — NVIDIA Newsroom, 2026-07-20 [blog]
- Integrate NVIDIA Omniverse RTX Sensor Simulation Into Existing Apps — NVIDIA Developer, 2026-07-20 [blog]
- NVIDIA NVLink: The Scale-Up Network for AI Factories — NVIDIA Developer, 2026-07-20 [blog]
- Bristol Myers Squibb Building Life Science Industry’s Most Advanced AI Factory on NVIDIA Vera Rubin — NVIDIA Blog, 2026-07-20 [blog]
- Gritt exits stealth with $34 million for robots to build solar plants — TechCrunch, 2026-07-21 [blog]
- X relaunches a rebuilt Android app after year-long effort — TechCrunch, 2026-07-20 [blog]
- Adobe camera app’s new feature will critique your photos using AI — TechCrunch, 2026-07-20 [blog]
- Claude disproves an 87-year-old math problem — Rundown, 2026-07-21 [blog]
- Agent swarms and the new model economics — Cursor, 2026-07-20 [blog]
- Boris has some really good advice — t3dotgg, 2026-07-21 [video]