Skip to content
Source · Daily Brief

AI Daily Brief — 27 July 2026

AI security set the tone as Nvidia and Microsoft launched an Open Secure AI Alliance that pointedly left out OpenAI, Google and Anthropic. Enterprise deals, a Chinese open-weight strategy and fresh copyright fights rounded out a busy day.

Top stories

  • Nvidia and Microsoft launch an Open Secure AI Alliance. The safety-and-security group formed without OpenAI, Google, or Anthropic. via The Verge – AI
  • Cognizant and Anthropic expand their partnership. The deal brings Claude to Cognizant’s enterprise clients. via Anthropic
  • OpenAI posts on AI at work. The company argued AI is expanding what people do on the job rather than only replacing tasks. via OpenAI
  • ChatGPT stops cloning famous writers’ voices. OpenAI curbed the behavior, though outputs may still capture a similar feeling. via Ars Technica – AI
  • The Hugging Face attack had precedents. A report pushed back on OpenAI’s framing of the incident as unprecedented. via MIT Technology Review – AI
  • Why China gives away its best AI models. An analysis weighed how Chinese open-weight releases affect US companies. via The Verge – AI
  • Meta’s models advance assistive robotics. Its SAM and DINO systems are helping the University of Pittsburgh transform assistive robotics. via Meta AI (scraped)
  • Artist sues an AI meme generator. The suit claims a deeply personal comic was sold as an ad template. via Ars Technica – AI

Who shipped

Anthropic widened its enterprise reach, teaming with Cognizant to put Claude in front of corporate clients. OpenAI was active on the communications front, publishing on AI’s changing role at work and revising policy so ChatGPT stops imitating named writers. Meta described deploying its SAM and DINO vision models for assistive-robotics research at the University of Pittsburgh. Microsoft unveiled AI security tools it says outperform rivals, while NVIDIA shipped agent-harness guidance and an Ising system for automated quantum-computer calibration. Cohere published a workflow study putting its Command model to work in wealth management.

Open-source pulse

The open-weight thread of the day was geopolitical: one analysis dug into why Chinese labs keep releasing their strongest models for free and what that means for US firms. On tooling, llama.cpp pushed a steady run of releases from b10144 through b10155. Research previews leaned open too, including Nanbeige4.2-3B, a compact model built for agentic tasks, and BiWM, an open-source interactive video world model.

Quiet corners

Google DeepMind surfaced only as an infrastructure counterparty, not as a model-maker, with no release of its own. Mistral, xAI and DeepSeek stayed silent across the day’s feed.

Money, infra & hardware

Verizon leaned into AI infrastructure with mini data centers and a $1 billion dark-fiber deal with Google. via Ars Technica – AI A web scraper’s court win against Google and Reddit kept the data-access fight in view, and Microsoft’s new security tooling arrived alongside the Nvidia-led alliance.

By the numbers

  • 130 stories across 19 sources
  • Most-mentioned model: GPT
  • Most-mentioned lab: NVIDIA
  • Notable absences: Mistral, xAI, DeepSeek

Compiled by AI Feed’s editor from all 130 headlines published on 27 July 2026.