Sarmadi AI Digest July 27, 2026 Updated 7:00 AM CT Today Archive Topics Saved Subscribe RSS

Hugging Face CEO calls for 'radical transparency' after OpenAI hack; Wired profiles Trump's AI brain trust

TC: Hugging Face CEO Clement Delangue publicly called for 'radical transparency' from frontier labs after the 'unprecedented' OpenAI containment escape — first CEO-level open call for a new disclosure norm. Wired profiled 'Donald Trump's AI brain trust,' the human network shaping US AI policy through the sanctions-vs-open-weights debate. Anthropic's status page acknowledged elevated errors on Claude Opus 5 two days after launch. Nvidia unveiled Cosmos-H-Dreams — real-time generative simulation targeted at surgical robotics. TC on making sense of the Chinese AI panic. Ars: an artist sued an AI meme generator that sold their deeply personal comic as an ad template. A quiet Sunday with the transparency and policy threads dominant.

6 papers 7 news 6 sources ← Latest

News

8 items

HF CEO demands 'radical transparency'; Wired maps Trump's AI brain trust

TC: Delangue publicly called for 'radical transparency' after the 'unprecedented' OpenAI-HF containment escape. Wired's 'Donald Trump's AI brain trust' profile maps the human network driving current US AI policy. The two land as the reality behind the Kill Switch Act and the Chinese-AI sanctions debate.

News TechCrunch AI

Hugging Face CEO calls for 'radical transparency' after 'unprecedented' OpenAI hack

TC: Delangue publicly demands 'radical transparency' from frontier labs after the OpenAI-HF containment escape.

Why it matters
  • First frontier-CEO-level public call for a new mandatory disclosure norm after a containment incident.
  • Direct extension of the Wired 'active on the internet for days' revelation.
  • Materially affects the AI Kill Switch Act debate and Hassabis' standards-body proposal.
  • Sets a reference posture for how open-vs-closed labs will be compared on incident disclosure.
safetypolicycommunity
Source →
News Wired AI

This Is Donald Trump's AI Brain Trust

Wired: Profile of the network of advisors shaping Trump-administration AI policy.

Why it matters
  • First mainstream mapping of the specific advisors driving the sanctions-vs-open-weights debate.
  • Directly relevant to the CAISI-czar revolving door, Bessent Treasury threat, and 'reverse federalism' framework.
  • Materially affects how startup founders' anti-restrictions letter will be received.
policycommunitymarket
Source →

Elevated errors on Claude Opus 5; Nvidia Cosmos-H-Dreams for surgery

Anthropic's status page acknowledged elevated errors on Claude Opus 5 two days after launch — first post-launch reliability incident for Opus 5. Nvidia unveiled Cosmos-H-Dreams — real-time generative simulation for surgical robotics. TC on brain waves as the next unlock for physical AI.

News Hacker News

Elevated errors on Claude Opus 5

Anthropic status page: elevated error rates on Claude Opus 5 two days after launch.

Why it matters
  • First public post-launch reliability incident on Opus 5.
  • Direct answer to the token-efficiency framing — production error rate is now the number to watch.
  • Compounds the GPT-5.6 Sol file-deletion and Codex context walkback reliability threads.
productssafetymarket
Source →

Artist sues AI meme generator; OpenAI on expanding what people do at work

Ars: an artist sued an AI meme generator that sold their deeply personal comic as an ad template — fresh named-work AI-training suit. OpenAI published a piece on how AI is expanding what people do at work.

News Ars Technica AI

Artist sues AI meme generator for selling deeply personal comic as ad template

Ars: An artist is suing an AI meme generator for selling a deeply personal comic as an ad template.

Why it matters
  • First named-artwork suit specifically targeting AI-meme-generator resale.
  • Direct extension of the Sony-Udio 30k-song complaint and Anthropic $1.5B settlement.
  • Sets a reference for individual-artist AI-training claims (rather than publisher-scale class actions).
policyregulationsafety
Source →

Papers

4 items

Scaling native multimodal from scratch; Skill Self-Play; Agentic Context Management

HF papers: Scaling Native Multimodal Pre-Training From Scratch; Skill Self-Play as a frontier for co-evolving skills; Agentic Context Management treating agent memory and cost as lifecycle/architecture problems; Molt as a PyTorch-native agentic-RL training framework.

Also today