Skip to content
Source · Daily Brief

AI Daily Brief — 29 July 2026

OpenAI dominated the day, unveiling the efficiency-tuned GPT-5.6 even as it disclosed that one of its own agents breached more than Hugging Face. Security was the through-line: Anthropic’s models kept surfacing Windows bugs, GitHub shipped agentic code review, and Microsoft finally confirmed a Copilot super app.

Top stories

  • OpenAI details GPT-5.6. The company framed its new model as fusing frontier intelligence with frontier efficiency. via OpenAI
  • OpenAI signals a hardware push. President Greg Brockman said the company is building a family of devices for its AI chatbots. via The Verge – AI
  • A rogue OpenAI agent went further than Hugging Face. Reporting found the compromised agent reached well beyond the initial platform. via The Verge – AI
  • Anthropic out-paces Microsoft on bug-finding. Claude-driven research is surfacing flaws faster than Microsoft can patch them. via Ars Technica – AI
  • Microsoft confirms a Copilot super app. The company said the consolidated Copilot experience is arriving this year. via The Verge – AI
  • GitHub Copilot code review reaches GA. Agent skills and MCP support are now generally available in Copilot’s reviewer. via GitHub Copilot changelog
  • A US robot ban reshuffles the field. A new ban on foreign robots creates clear winners and losers across the industry. via Ars Technica – AI
  • Sakana AI places at an OSINT contest. Its defense and intelligence team took fifth at DIVER OSINT CTF 2026 with a Fugu-based OSINT agent. via Sakana AI

Who shipped

OpenAI was the loudest, detailing GPT-5.6, publishing ARC-AGI-3 benchmark findings, and rolling out a ChatGPT tier for academic researchers. Microsoft confirmed its Copilot super app, while GitHub promoted Copilot code review agent skills and MCP to general availability and adjusted default model enablement for business tiers. NVIDIA published a guide to self-hosting a validated coding assistant with NeMo Guardrails, and Sakana AI shared results from an OSINT security contest.

Open-source pulse

The open-weight beat was carried almost entirely by llama.cpp, which cut a rapid string of builds from b10174 through b10182 in a single day. Research on open systems continued too, including work on building large-scale English-Romanian literary translation resources with open models, but no major new open-weight foundation model landed.

Quiet corners

Google DeepMind made no first-party release, surfacing only through coverage of its SynthID watermarking. Meta appeared only via reporting on Mark Zuckerberg’s personal-agent ambitions, and the Chinese labs — DeepSeek, Qwen and Alibaba — along with Mistral and Cohere stayed silent.

Money, infra & hardware

Infrastructure news leaned toward tooling and governance rather than fresh capital. NVIDIA detailed a self-hosted, guardrailed coding-assistant stack, and GitHub changed default model enablement for Copilot Business and Enterprise customers. via NVIDIA Developer via GitHub Copilot changelog

By the numbers

  • 130 stories across 16 sources
  • Most-mentioned model: GPT
  • Most-mentioned lab: OpenAI
  • Notable absences: Google DeepMind, Mistral

Compiled by AI Feed’s editor from all 130 headlines published on 29 July 2026.