AI News for September 2, 2026 — 7min.ai

[NEWS] Anthropic ships Claude Fable 5.1 and Mythos 5.1, cutting agentic costs up to 45%: Anthropic released Claude Fable 5.1 and Mythos 5.1, the same underlying model split by safeguard level.
[NEWS] OpenAI says Astra is its first model to cross the Critical cybersecurity threshold: OpenAI disclosed that Astra, its forthcoming model suite, is the first to meet the Critical cybersecurity capability threshold under its Preparedness Framework.
[RESEARCH] Runway's Solaris renders apps as live video with no code underneath: Runway previewed Solaris, an 'Interface World Model' that draws websites and apps frame by frame as real-time video, responding directly to clicks, drags, and voice.
[NEWS] ChatGPT Health connects to Epic, opening records for 325M patients: OpenAI integrated ChatGPT Health with Epic's electronic health record system, letting clinicians import patient data and query it conversationally.
[NEWS] Google joins Nvidia's $3.5B MediaTek chip deal, shares jump 10%: Google is also investing alongside Nvidia in the $3.5 billion AI chip deal with Taiwanese chipmaker MediaTek, a detail not disclosed when Nvidia's investment first surfaced.
[NEWS] Anthropic reverses course on data retention with Enterprise Frontier Safeguards: Anthropic is rolling out Enterprise Frontier Safeguards, a system that gives customers the equivalent of zero data retention while still monitoring for adversarial use.
[NEWS] Nvidia's Hugging Face deal price climbs to $14B, close expected this week: Nvidia is now in advanced talks to acquire Hugging Face for about $14 billion, roughly $1 billion above the $12.9 billion price reported when the deal first surfaced.
[NEWS] Cognition raises $1B at a $47B valuation: AI coding startup Cognition is closing a funding round of roughly $1 billion that values it at about $47 billion.
[TOOLS] DeepMind adds agentic video understanding to Gemini Flash models: Google DeepMind launched agentic video understanding across Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite.
[RESEARCH] Google Research: models can recover 65% of facts they fail to recall, just by thinking longer: A study from Google Research and Technion found that frontier models encode 95-98% of tested facts in their weights but fail to surface them during generation.

About 7min.ai · Subscribe to newsletter