Tuesday, 1 September 2026
CrowdStrike ships AI that attacks itself
Archive item — written before sources were shown.
CrowdStrike's SafeMind pairs an offensive red-team model with a defensive one in the same loop, trained on 15 years of its own incident data.
World Labs ships an omni world model
Archive item — written before sources were shown.
Atlas generates and reconstructs 3D scenes from text, images, video, and depth maps in one model, pretrained from scratch rather than bolted onto an LLM.
Anthropic ships Fable 5.1 and Mythos 5.1
Archive item — written before sources were shown.
Anthropic's point release cuts agentic-work costs up to 45% and adds a diagnostic case study Millennium's own engineers couldn't crack for years.
AIR raises $50M to vet AI agent plugins
Archive item — written before sources were shown.
Sequoia and Greenoaks backed AIR's exit from stealth, a real-time system that discovers which AI agents run inside a company and vets their skills and plugins.
Anthropic lets enterprises hold their own Claude data
Archive item — written before sources were shown.
Enterprise Frontier Safeguards moves Claude data into a customer's own AWS, Azure, or Google Cloud account and routes misuse alerts straight to the customer.
Anthropic signs $35B compute deal with Lambda
Archive item — written before sources were shown.
The six-year Nvidia-backed cloud agreement, built on a Texas site run by a former bitcoin miner, pushes Anthropic's compute contracts past $135B for the year.
OpenAI's own team doubled data coverage with Clay
Archive item — written before sources were shown.
OpenAI's revenue-ops team used Clay's AI agents to take data-enrichment coverage from 40% to 80%, one of three case studies on AI-native workflows.
Japan's QommonsAI now covers 1,050 municipalities
Archive item — written before sources were shown.
Polimill built its public-sector AI platform on OpenAI's models; it now serves roughly 550,000 government employees across Japanese local government.