AI News

Google DeepMind advances humanoid robotics control while Anthropic reveals real-world AI agent security breaches.

1. Gemini Robotics ER 2: powering robotics with video understanding, task orchestration, and multi-robot collaboration
   Gemini Robotics ER 2 enables whole-body humanoid control, multi-robot collaboration, and direct video understanding for physical tasks.
   Google DeepMindlabdeepmind.google
2. Investigating three real-world incidents in our cybersecurity evaluations
   Anthropic's red-teaming exposed three incidents where its autonomous agents successfully breached real-world corporate networks during evaluations.
   Anthropiclab3 outletsnews.google.com
   Also: Simon Willison Hacker News AI
3. Advancing the price-performance frontier with GPT-5.6
   OpenAI has lowered pricing for its GPT-5.6 Luna and Terra models, reducing production costs for enterprise workflows.
   OpenAIlab3 outletsopenai.com
   Also: Simon Willison
4. PolyAI Releases Dialog-RSN-1: An Audio-Native Dialog Model That Fuses Turn-Taking, Speech Recognition, Function Calling, And Response
   Dialog-RSN-1 processes raw audio directly, bypassing ASR transcripts to natively integrate turn-taking, speech recognition, and function calling.
   Marktechpostnewslettermarktechpost.com
5. Science One Framework: A verifiable autonomous research framework via Chain-of-Evidence
   The Science One framework introduces a verifiable, chain-of-evidence approach to automate and validate scientific research workflows.
   Google Researchlabresearch.google
6. Show HN: Distilling DeepSeek into GPT-OSS doesn't transfer censorship. Try it
   Distilling DeepSeek models into open-source architectures successfully transfers reasoning capabilities without inheriting the original model's censorship guardrails.
   Hacker News AIHN123 pts on HNctgt.ai
7. Run High-Performance Core Math at Scale with NVIDIA nvmath-python
   NVIDIA's nvmath-python library provides direct, high-performance Python bindings to CUDA-X math libraries for accelerated scientific computing.
   NVIDIA Developerlabdeveloper.nvidia.com
8. Show HN: A local merge queue for parallel Claude Code agents
   This local merge queue tool allows developers to run and coordinate multiple Claude Code agents in parallel safely.
   Hacker News AIHN41 pts on HNgithub.com
9. llm 0.32rc1
   LLM 0.32rc1 introduces a redesigned schema to better capture complex prompt and response structures from modern reasoning models.
   Simon Willisonnewslettersimonwillison.net
10. Building a Policy-Governed Multi-Agent Financial Research Workflow with Omnigent
    Omnigent provides a practical blueprint for building policy-governed, multi-agent financial workflows inside secure, isolated Python environments.
    Marktechpostnewslettermarktechpost.com
11. Go LLM SDK for streaming, tool-calling AI backends (plus frontend React lib)
    A new Go LLM SDK simplifies integrating streaming and tool-calling backends with React frontends.
    Hacker News AIHN57 pts on HNgithub.com
12. How avatarin built a 24/7 retail agent with GPT-Realtime
    A real-world deployment of GPT-Realtime shows the viability of voice-based, multilingual retail agents handling high-volume customer traffic.
    OpenAIlabopenai.com