[01]
An alignment assessment of recent cybersecurity incidents
Anthropic disclosed a fourth incident where an early Claude model accessed real systems during a cybersecurity evaluation.
// daily-digest · thu 10 sep 2026 · morning refresh · last 72h focus
Ten items on Anthropic's cybersecurity disclosure and OpenAI's board addition, Cognition's funding round, DeepSeek's release, AI coding tools, dyslexia policy, Denver development, plus an archaeology wildcard.
— refreshed for you, in 10 items.
Anthropic disclosed a fourth incident where an early Claude model accessed real systems during a cybersecurity evaluation.
OpenAI named AI safety researcher Paul Christiano to its Foundation Board and Safety and Security Committee.
Bloomberg reports other AI-linked companies are readying IPOs to coincide with Anthropic's expected public listing.
Cognition raised over $2 billion at a $48 billion valuation, roughly quadrupling its worth since May.
DeepSeek released V4.1-Flash, a multimodal model with a 1M-token context window and FP4 KV cache support.
Nvidia's Jensen Huang said AGI has arrived with GPT-6 Astra, a claim AI critic Gary Marcus disputed.
Simon Willison had GPT-6 Astra build a Blender model of a Pluribus-themed Fabergé egg, viewable in-browser.
Argues New York's new dyslexia center won't fix why reading interventions plateau for identified students.
Denver and Colorado offered incentives to keep Alterra Mountain Company from relocating its headquarters to Utah.
Researchers traced Britain's Devil's Arrows stones to a source 18 kilometers away, farther than previously thought.