Skip to content
Source · Daily Brief

AI Daily Brief — 1 February 2026

February opens on a Sunday — quiet in primary press releases but loaded with structural signals. Jack Clark‘s Import AI #443 goes out anchored to “Sunday February 1st.” The Colorado AI Act‘s original effective date passes without taking effect (now deferred to 30 June). The EU AI Act‘s one-year prohibited-practices anniversary lands tomorrow, triggering the European Commission’s first mandatory review. The International AI Safety Report 2026 — Bengio-led, 100+ experts, 30+ countries — is queued for Tuesday release.

Top stories

  • Import AI #443 — agent ecologies and the internet in transition. Jack Clark’s editorial opens with Moltbook and OpenClaw as agent-society thought experiments; the issue surveys three arxiv preprints (Fauna Robotics’ Sprout developer-ready humanoid; Skoltech’s Glove2UAV wearable IMU-based drone control; and Bosse et al.’s Automating Forecasting Question Generation and Resolution for AI Evaluation). via Import AI
  • CSET — “When AI Builds AI.” A multi-org workshop report (Helen Toner, Steve Newman, Ryan Greenblatt, Peter Wildeford, Samuel Albanie) argues that scenarios are possible in which AI R&D becomes highly automated, accelerates dramatically and the resulting systems pose extreme risks. Surfaced in the Feb 1 weekend window. via CSET / Georgetown
  • Anthropic Engineering — “Designing AI-resistant technical evaluations.” Tristan Hume walks through three iterations of Anthropic’s take-home performance test as successive Claude versions defeated each design — a concrete data point on capability uplift and eval obsolescence. via Anthropic Engineering
  • Colorado AI Act original effective date passes without taking effect. The first comprehensive US state AI law was scheduled to enter enforcement 1 February. Governor Polis signed SB 25B-004 last August postponing implementation to 30 June 2026 after a special legislative session failed to reach amendment compromise. The law will later be repealed and replaced by SB 26-189 in May. via Akin Gump

Money & infra

The week’s running infra deals stay live. NVIDIA‘s additional $2B investment in CoreWeave (announced 26 January at $87.20/share) commits to more than 5 GW of AI-factory buildout by 2030, with CoreWeave first to deploy NVIDIA’s forthcoming storage systems and new CPU. Cerebras closes the week with its $20B+ Master Relationship Agreement with OpenAI (750 MW of inference through 2028) heading into its imminent $1B Series H at $23B post-money — round officially announced Tuesday. Stargate Abilene begins bringing first GB200 racks online. via NVIDIA

Open-source & China

Moonshot Kimi K2.5 remains the freshest open-weight headliner from late January — a 1T-parameter MoE (32B active) trained on ~15T mixed visual+text tokens, with Agent Swarm orchestrating up to 100 sub-agents in parallel (~4.5× faster than single-agent on wide-search tasks). Z.ai‘s GLM-4.7-Flash (30B-A3B MoE, MIT-licensed, released 19 January) is the standout lightweight model trending on Hugging Face. CNBC’s running framing for the weekend: Chinese labs (Alibaba, Moonshot, MiniMax, Zhipu, ByteDance) are pre-positioning ahead of Lunar New Year (Year of the Horse, 17 February). DeepSeek V4 is the dominant pre-holiday anticipation story — though the eventual drop slips to 24 April. via Simon Willison

Policy & courts

Tomorrow’s EU AI Act one-year anniversary triggers the European Commission’s first mandatory review of banned practices — social scoring, real-time biometric ID in public spaces, emotion recognition in workplaces and schools, and untargeted facial-image scraping. Non-compliance penalties run up to €35M or 7% of global turnover. The International AI Safety Report 2026 drops Tuesday, with Yoshua Bengio leading 100+ AI experts backed by 30+ countries and international organisations. The Information reported Tuesday that Anthropic lifted its 2026 revenue forecast ~20% to $18B (and $55B for 2027). via European Commission

Quiet corners

No first-party shipping from OpenAI, Google DeepMind, Meta AI or Microsoft AI on Sunday. xAI‘s last move was the Wednesday Grok Imagine video API at $0.05/second. Thinking Machines Lab continues in stealth — January departures of Barret Zoph and Luke Metz back to OpenAI are the active narrative thread.

By the numbers

  • Anthropic 2026 revenue forecast: $18B (+20% from prior) · 2027: $55B
  • NVIDIA × CoreWeave: $2B investment · 5+ GW AI-factory buildout by 2030
  • Cerebras × OpenAI: $20B+ Master Agreement · 750 MW inference through 2028
  • Kimi K2.5: 1T params · 32B active · ~15T training tokens · Agent Swarm up to 100 sub-agents
  • EU AI Act fines: up to €35M or 7% of global turnover

Compiled by AI Feed’s editor from publicly reported announcements on 1 February 2026.