AI Daily Brief — 5 June 2026

Anthropic Science detailed how Claude is becoming a chemist, Perplexity opened Nemotron 3 Ultra to Pro and Max subscribers, and MIT Technology Review argued the Meta hack shows there”s more to AI security than Mythos. Tencent Hunyuan published planning research, and Nikolay Savinov joined OpenAI”s new London pretraining team.

Top stories

• Anthropic Science: Making Claude a chemist. Molecule-manipulation work — anchoring chemistry around a model that understands the manipulation space. via @AnthropicAI
• Nemotron 3 Ultra is now available for Pro and Max on Perplexity / Computer. NVIDIA”s frontier-grade model lands in a major consumer surface. via Perplexity
• Latent Space: How to Stop Shipping Low-Quality RL Environments. Practitioner-focused guide on environment quality for agent training. via Latent Space
• MIT Technology Review: The Meta hack shows there”s more to AI security than Mythos. Frames the recent breach as evidence the security frontier extends beyond capability gating. via MIT Technology Review
• Nikolay Savinov joins OpenAI in London to work on pretraining. A meaningful hire amplified by Altman. via @sama

Who shipped

Meta”s SAM 3D team picked up a Best Paper Honorable Mention at CVPR 2026. Tencent Hunyuan dropped a planning-focused LLM research paper. Google published its full May 2026 AI announcements digest.

Quiet corners

Mistral, DeepSeek were absent.

By the numbers

• 17 stories tracked across 13+ sources
• Most-mentioned model: Nemotron 3 Ultra
• Most-mentioned lab: Anthropic
• Notable absences: Mistral, DeepSeek

Compiled by AI Feed”s editor from all 17 headlines published on 5 June 2026.