AI News from October 30, 2026

Tyler Shultz says AI startups create new fraud risks because investors may struggle to distinguish genuinely novel systems from lightly packaged existing tools.

Reports say Chinese AI labs are shifting distillation efforts from Anthropic’s Claude to OpenAI’s Astra, raising new questions about model theft defenses.

DoorDash says third-party AI agents have generated only several hundred orders, underscoring how early agent-led commerce remains despite industry hype.

Epoch AI researchers say open-weight model GLM 5.3 built Firefox exploit chains autonomously, exposing patch-gap risks for browsers and software vendors.

DeepSeek reportedly raised 105.5 billion yuan at an $86 billion valuation, with Tencent and CATL leading as state-linked capital reaches 27%.

Wabi raised $55 million at a $1.25 billion valuation as it pivots from prompt-built apps to a personal AI agent for everyday tasks.

Dream has raised $17 million from Andreessen Horowitz to build physics-aware AI for construction, targeting design and engineering workflows in AEC.

NVIDIA is showing how AI agents and Omniverse libraries can build, test and optimize simulation workflows for robotics, factories and industrial design.

Mistral has released downloadable Le Chonk, while ZDNET places the open coding model near Codex GPT-6.1 Terra amid limited public evidence.