This brief covers AI news from 2026-09-05 UTC.
Today’s brief covers new AI launches, funding, and a key legal filing.
CrowdStrike unveils SafeMind security models with Nvidia
What happened: CrowdStrike introduced SafeMind, a family of purpose-built security models and harnesses, built with Nvidia Nemotron and trained on Falcon telemetry. It includes Red Tempest (offensive) and Blue Solano (defensive) models.
Why it matters: Security teams get an agentic system that automates attack path finding and defense, potentially reducing false positives and improving response.
GitHub’s HydraFusion router cuts coding costs up to 67%
What happened: GitHub announced Project HydraFusion, a research preview in Copilot that dynamically routes coding tasks across models. In tests, it beat Claude Opus 5 on TerminalBench 2.1 by 4.9 points while costing 67% less.
Why it matters: Developers can get frontier-level coding quality at a fraction of the cost by letting a router pick the best model per task.
DOJ tells court AI training is fair use in OpenAI case
What happened: The U.S. Department of Justice filed a statement of interest in the New York Times v. OpenAI lawsuit, arguing that training LLMs on copyrighted works is transformative fair use and that requiring licensing would hamper competition.
Why it matters: A ruling favoring fair use would keep training costs low for AI builders, while an adverse ruling could create licensing barriers favoring big tech.
Source: Publishing Perspectives
Nscale commits $3.5B to Figure AI for humanoid compute
What happened: AI cloud provider Nscale signed a multi-year agreement to supply at least $3.5 billion in GPU infrastructure to Figure AI, with plans to scale beyond $6 billion, and took an undisclosed equity stake.
Why it matters: Humanoid robotics training requires massive compute; this deal secures infrastructure and signals new financing models for physical AI.
Ant Ling releases visual agent model Ling-3.0-flash-VL
What happened: Ant Ling released Ling-3.0-flash-VL, a version of its language model that accepts images and video, can generate websites from screenshots, and operate GUIs via its hosted API.
Why it matters: Developers can build agents that see and interact with software, enabling automated UI testing and visual coding workflows.
SoundHound closes LivePerson acquisition, appoints CFO
What happened: SoundHound AI completed its acquisition of LivePerson, expanding its customer base to 25 of the Fortune 100 and adding over 750 patents. John Collins was appointed CFO of the combined company.
Why it matters: The combined platform integrates voice and digital messaging, giving enterprises a unified conversational AI system across channels.
Gimlet Labs raises $300M at $3B for multi-chip AI
What happened: Gimlet Labs raised $300 million at a $3 billion valuation, led by Andreessen Horowitz, with new backers Arm and Microsoft’s M12. Its software routes AI workloads across different chip types.
Why it matters: As GPU scarcity bites, multi-silicon orchestration can cut inference costs and speed up workloads by using the best chip for each task.