Lab Watch
Fast, clear reactions to major AI lab announcements, focused on real-world impact and opportunity.
Updates
Google DeepMind Ships Gemini 3.8 Flash and a Restricted Cybersecurity Variant
Google’s third Flash update in six weeks adds a gated Gemini 3.8 Flash Cyber model for vetted defenders through a new Fairwind Program.
Meta Releases Muse Spark 1.3, Its Most Capable Coding and Agent Model Yet
Meta says its new closed-weight model now competes with Anthropic’s and beats OpenAI’s flagship on coding, while using fewer tokens per task.
OpenAI commits $1 billion to give under-resourced cyber defenders access to its Daybreak tools
OpenAI is subsidizing frontier cybersecurity AI for water utilities, power grids, small banks, and local governments over the next six months.
Anthropic open-sources a commerce-agent blueprint for Claude, with Visa, Mastercard, and Shopify already building on it
Anthropic released free reference code for shopping and merchant AI agents, backed by Visa, Mastercard, Accenture, and Shopify.
Anthropic Launches Claude Fable 5.1 and Mythos 5.1, Cutting Cache Prices 75%
Anthropic’s newest Claude models top its benchmark charts and slash agentic workload costs, with a more permissive twin for vetted defenders.
OpenAI Releases GPT-6 Astra, Its First Model Rated ‘Critical’ for Cyber Risk
OpenAI’s new flagship model, GPT-6 Astra, ships with unprecedented cybersecurity safeguards after a summer AI containment breach.
Agent Horizon 