← archive · dailyblip

Tuesday, July 21 — six things that changed today

Issue 244 · 2026-07-21

The brief

1. OpenAI's CEO is lobbying to restrict open-weight AI models — a direct policy signal that could reshape which tools creators are legally allowed to use.
2. Sony Music is suing Udio, alleging 30,000 copyrighted songs — Elvis to Beyoncé — were ingested without permission to train its AI music generator.
3. A federal judge gave final approval to Anthropic's $1.5B copyright settlement — the largest AI training liability deal yet, though broader training questions remain open.
4. Google's new "Frozen v2" chip targets 6–10x Gemini inference efficiency — if it ships in 2028, expect faster Veo and Imagen generation limits to follow.
5. AI ingested 40 million complete recordings where music law only ever grappled with four bars — the scale of the legal gap facing music creators is now undeniable.
6. The Model Context Protocol shipped a stateless session update simplifying how AI tools connect — if you build or use MCP-based workflows, check your stack now.

Tool drop

LayerProof Mylar

Generates social content for creators — turns your ideas into ready-to-post assets without manual design work.

The feed that day

Anthropic's $1.5B copyright settlement approved by federal judge

A federal judge gave final approval to Anthropic's landmark settlement with authors and publishers over illegal training on copyrighted books. The ruling doesn't resolve the broader question of AI training on copyrighted works.

rights · via TechCrunch AI

Sony sues Udio over 30,000 copyrighted songs in AI music generator

Sony Music filed a copyright lawsuit alleging Udio ingested over 30,000 of its songs without permission, from Elvis to Beyoncé.

rights · via The Verge AI

Google developing new AI chip for more efficient Gemini models

Alphabet is designing "Frozen v2," a server chip targeting 6–10x efficiency gains in token generation per watt, expected 2028.

industry · via TechCrunch AI

Model Context Protocol gets stateless session update for easier integration

The MCP, core infrastructure for AI model interoperability, is shipping a significant update making server-side session management simpler, similar to standard web architecture.

tools · via TechCrunch AI

OpenAI head pushes for US regulation against open-weight models

Strategic debate over Chinese open-weight LLMs, OpenAI's regulatory concerns, and pushback from industry figures on innovation. Direct policy signal for AI creators.

industry · via TechCrunch AI

The Sample Clearance Problem, Except It's the Whole Song Now

The music industry spent three decades building a legal framework to stop people from lifting four bars of someone else's record. AI just swallowed 40 million complete recordings and called it research.

writing · via dailyblip

Scientists warn AI-edited wildlife photos could compromise research databases

Researchers caution that AI-enhanced bird photos shared on citizen science platforms may undermine credibility of iNaturalist and Macaulay Library. Policy/ethics signal.

rights · via PetaPixel

Adobe Project Indigo launches AI Playground for mobile photography

Project Indigo now includes generative AI for photo critiques, object removal, depth of field, and style transfer. Overlaps with items 2 and 3; PetaPixel version.

image · via PetaPixel

Adobe's Project Indigo integrates generative AI into iPhone photography

Project Indigo now includes generative AI tools powered by Google's Nano model for editing, removal, and style application. Overlaps with item 2; uses different framing.

image · via The Verge AI

NVIDIA releases Cosmos 3 Edge for edge-based world modeling

NVIDIA shipped Cosmos 3 Edge, an AI model for physical systems to understand scene changes and anticipate outcomes on memory-constrained hardware. Released tool.

tools · via Hugging Face Blog

Adobe Project Indigo adds LLM photo critique and generative features

Adobe's experimental iOS camera app now uses LLMs to critique photos and includes AI-powered object removal, depth of field, and style transfer. Shipped feature.

tools · via TechCrunch AI

Lightcraft's Spark Story uses iPhone to generate virtual film sets from scripts

SIGGRAPH 2026 revealed Spark Story, which converts screenplays into USD/Gaussian Splat environments and enables on-device shot planning before production.

tools · via Creative Bloq AI

YouTube clarifies monetization policy for AI-generated and low-quality content

YouTube updated policies defining which AI-generated and repetitive videos can't earn ad revenue, adding nuance to existing inauthentic-content rules.

rights · via TechCrunch AI

Moonshot and Alibaba release frontier AI models challenging OpenAI and Anthropic

Chinese AI companies launched competitive models claimed to match OpenAI/Anthropic capabilities at lower cost, narrowing the US AI lead.

industry · via The Verge AI

OpenAI discusses safety lessons from long-horizon AI deployments

OpenAI shares findings on safety risks and safeguards from running extended-duration AI models. Policy and research signal.

industry · via OpenAI News

DeepMind AI reconstructs Pelé's legendary unfilmed goal using motion synthesis

Google DeepMind recreated a famous 1959 goal Pelé claimed was his best but was never captured on video, using AI to synthesize the play from his account.

video · via PetaPixel