Sent to 23 readers
Race to AGI Daily Digest - Wednesday, July 22, 2026
Share:
TLDR
Anthropic’s $1.5bn copyright settlement sets a new benchmark for the cost of training data in frontier AI.
WAIC 2026 launches global agent safety standards and a cooperation bloc, formalizing how agents should be evaluated and governed.
Agentic AI phones, robots, and edge devices showcased at WAIC push embodied and on-device AI further out of the lab.
OpenAI balances politics and builders, with a Trump-era briefing on frontier models and a live Codex-building stream for creators.
Chipmakers Intel, AMD, ARM, and TSMC surge as investors double down on the hardware backplane for agentic and scientific AI.
The Full Story
Following Monday’s open-source Mixture-of-Experts push from Tencent and Tuesday’s $450M bet on CuspAI’s materials foundry, today the camera pulls back. The same GPUs and models are in play, but the story is about rules, rights, and who gets to steer the stack.
Start with Anthropic. A $1.5bn copyright settlement doesn’t just clean up a lawsuit; it effectively puts a price list on training data for frontier models. Anthropic settlement details -> Anyone racing in the NVIDIA–Groq era now has to factor in not just chips and power, but long-tail content liability.
Over in Shanghai, WAIC is giving shape to two of our running threads. On one side, you’ve got agentic AI phones, robots, and edge boxes, with Siemens and others showing how world models and local inference spill into factories and devices. WAIC’s agentic AI showcase -> On the other, a new bloc is pushing global agent safety standards, turning “don’t let the agent go rogue” into checklists and audits. Global agent safety standards summary ->
OpenAI sits right at this intersection. Sam Altman is heading to brief Trump officials on next‑gen frontier models, while OpenAI Academy lines up a live Codex-building stream to pull more creators into the agent world. Altman briefing coverage -> OpenAI Academy Codex livestream -> At the same time, Hacker News is digging through a security incident involving OpenAI and Hugging Face during model evaluation, a reminder that “agentic AI meets enterprise risk” isn’t abstract. HN security incident thread ->
Meanwhile, the market is quietly voting for more pipes. Intel, AMD, ARM, TSMC, and Oracle all rallied hard today, reinforcing our hardware-supply-chain storyline even as policy dominates the headlines. The week started with open models and scientific discovery; by midweek, the message is clear: the race to AGI now runs through courtrooms, standards bodies, and chip fabs at the same time.
Get This Delivered Daily
Join thousands of AI professionals who start their day with Race to AGI.