Skip to content
Source · Daily Brief

AI Daily Brief — 1 August 2025

A regulatory-charged Friday closing one era and opening another. OpenAI cleaned up a privacy spill. Brussels published GPAI signatories. Mistral filled out a $1B round. Beijing slapped back on H20.

Top stories

  • OpenAI kills ChatGPT “make chats discoverable” feature. After Fast Company reported nearly 4,500 conversations — some with names, locations, mental-health and business details — surfaced on Google, OpenAI CISO Dane Stuckey announced removal. Called it a “short-lived experiment” that “introduced too many opportunities for folks to accidentally share things they didn’t intend to”. Working with Google to de-index existing links. via Fortune
  • EU AI Office publishes GPAI Code of Practice signatories. One day before GPAI obligations under Articles 53 & 55 take effect on Aug 2. The Code — Transparency, Copyright, and Safety & Security chapters — was endorsed by the AI Board as an adequate compliance tool, with signatories gaining reduced administrative burden vs non-signatories. via European Commission
  • Mistral in talks to raise $1B at $10B valuation. FT reports Mistral AI in late-stage talks with investors — reportedly including Abu Dhabi-backed MGX — for $1B at ~$10B valuation, nearly double the prior €5.8B mark. Funds for compute and frontier-model training. Round eventually closed in September at €12B ($14B) with $1.5B anchor check from ASML. via PYMNTS / FT
  • China CAC summons NVIDIA over alleged H20 “backdoor” risks. Beijing reportedly tells ByteDance, Alibaba, Tencent to halt H20 orders pending security review — only weeks after Trump administration cleared NVIDIA to resume sales. NVIDIA would later (Aug 22) reportedly move to halt H20 production in response. via CNBC
  • Last day before EU AI Act GPAI obligations enter force. From Aug 2, GPAI providers must produce technical documentation, supply downstream-deployer information, publish a training-data summary using the AI Office template, and adopt an EU-copyright compliance policy. Existing models get until Aug 2, 2027. AI Office gains full enforcement powers Aug 2, 2026. via Baker McKenzie

Who shipped

OpenAI shipped a privacy cleanup. Brussels shipped a signatories list. Mistral shipped a funding term sheet.

Open-source pulse

OpenAI in final ramp to gpt-oss — first open-weight release since GPT-2. Aug 5 drop staged under Apache 2.0: 120B MoE matches o4-mini on core reasoning, fits on single 80GB H100; 20B targets o3-mini on 16GB edge device. Both expose low/medium/high reasoning-effort knobs. Anthropic staged Claude Opus 4.1 for Aug 5 across Bedrock, Vertex and GitHub Copilot (74.5% on SWE-bench Verified vs 72.5% for Opus 4).

Money, infra & hardware

Black Forest Labs and Microsoft prepared Aug 5 launch making FLUX.1 Kontext [pro] and FLUX1.1 [pro] available directly inside Azure AI Foundry — BFL’s biggest enterprise channel push to date. Sacra estimated BFL at ~$96M annualised revenue. Fal’s $125M Series C from July 31 at $1.5B still the topline gen-AI infra story.

Quiet corners

Anthropic disclosed Project Fetch — Claude vs unaided humans on a quadruped robot. Eight non-roboticist researchers split into teams; Claude-assisted team finished in roughly half the time and made the only substantial progress toward fully autonomous retrieval. via Anthropic Trump AI Action Plan fallout — agencies scoping “preventing woke AI” procurement guidance.

By the numbers

  • ~4,500 — private ChatGPT conversations indexed on Google before takedown
  • ~$1B / $10B — Mistral round target / valuation
  • Aug 2 / Aug 2 2027 / Aug 2 2026 — GPAI effective / legacy grace / fines effective
  • 120B / 20B — gpt-oss MoE sizes (Apache 2.0)
  • 74.5% vs 72.5% — Opus 4.1 vs Opus 4 on SWE-bench Verified

Compiled by AI Feed’s editor from verified web sources for 1 August 2025.