Skip to content
Source · Daily Brief

AI Daily Brief — 3 February 2026

Tuesday delivers the biggest single day of the AI-safety year so far. The International AI Safety Report 2026 — led by Yoshua Bengio, written by more than 100 AI experts and backed by 30+ countries and international organisations — publishes, intended to inform the India AI Impact Summit and ongoing regulatory work worldwide. Cerebras officially closes its $1B Series H at a ~$23B post-money. Anthropic moves on two non-trivial fronts: life-sciences partnerships with the Allen Institute and HHMI, and selection by the UK government to build and pilot a GOV.UK AI assistant. Google DeepMind ships Project Genie — a world-model prototype letting users create and explore generated 3D worlds in real time.

Top stories

  • International AI Safety Report 2026. The second edition of the report — the largest global collaboration on AI safety to date — lays out a science-based assessment of general-purpose AI capabilities, risks (misuse, systemic, control), and mitigations. Authored by 100+ experts; backed by 30+ countries and international organisations; intended as the working input to the India AI Impact Summit. via International AI Safety Report
  • Cerebras — $1B Series H at $23B. Tiger Global leads, with Benchmark, Fidelity, Atreides, Alpha Wave Global, Altimeter, AMD, Coatue and 1789 Capital participating. The round comes four months after the September 2025 Series G ($1.1B at $8.1B) and two weeks after the $10B Master Relationship Agreement with OpenAI (750 MW of inference compute by 2028). The valuation roughly triples in four months. via Cerebras
  • Anthropic — Allen Institute + HHMI partnership. Two life-sciences founding partners: HHMI gets specialised Claude-powered lab agents integrated with scientific instruments and analysis pipelines; the Allen Institute will co-develop multi-agent systems for multimodal data analysis. via Anthropic
  • Anthropic — GOV.UK pilot. The UK government selects Anthropic to build and pilot a Claude-powered AI assistant for GOV.UK digital services — a non-trivial sovereign-AI win against US competitors. via Anthropic
  • Google DeepMind — Project Genie. An experimental research prototype lets users create and explore generated 3D worlds in real time. Rolls out to Google AI Ultra subscribers. via Google

Generative media & robotics

ElevenLabs v3 reaches general availability — pitched as the company’s most advanced text-to-speech to date. Figure unveils Helix 02, extending its single neural network to full-body control across the humanoid platform. NVIDIA publishes Cosmos Policy on Hugging Face for general-purpose robot control and manipulation. Three robotics / voice ships in one day. via ElevenLabs

Quiet corners

OpenAI ships only a minor change — another small Standard thinking-time tweak for GPT-5.2 (follow-up to the 10 January cut). Meta AI, xAI, Microsoft AI and Thinking Machines Lab are silent on first-party shipping. The previous day’s Snowflake × OpenAI $200M partnership and EU AI Act one-year anniversary keep trending into Tuesday coverage.

By the numbers

  • International AI Safety Report 2026: 100+ AI experts · 30+ countries and international organisations
  • Cerebras Series H: $1B raise · ~$23B post-money · valuation ~3× in four months
  • Cerebras × OpenAI commitment: $10B · 750 MW by 2028
  • Three robotics + voice ships: ElevenLabs v3 GA · Figure Helix 02 · NVIDIA Cosmos Policy

Compiled by AI Feed’s editor from publicly reported announcements on 3 February 2026.