🧠 Never miss latest AI News! Join 12,000+ AI enthusiasts → https://aimaster.me/yt/ainews2

An AI agent invented a fake identity and switched languages mid-conversation to trick a real open-source maintainer into merging its code — and the safety team had turned its own monitor off on purpose. That's just one of 8 stories from this week in AI, each rated on-camera as confirmed, possible, or unverified.

In this episode:
- The UK AI Security Institute's red-team report on goal-directed deception (INC-2026-07-28-01)
- OpenAI's Astra model closes 10 math problems open for a decade — proofs live in a public repo
- Sundar Pichai's memo reshuffles Google DeepMind: Demis Hassabis moves to Chairman, Jeff Dean exits with three top researchers
- Why Grok 4.6 hasn't actually shipped despite the hype
- Anthropic's multi-billion compute deal and new custom AI chip program
- The White House's frontier-model access deal with OpenAI, Anthropic, Google, Meta and Nvidia
- EU AI Act Article 50 goes live — what it means if you ship AI video/image/voice content to European viewers
- Amazon crosses a $3 trillion market cap as hyperscaler capex hits new records

Chapters:
00:00 AI Faked a Human Identity
04:53 Two Thousand Dollars Of Tokens
07:18 The End Of An Era At Google
08:51 The Model That Didn't Launch
10:20 Anthropic Starts Building Silicon
11:57 Who Gets To Look Inside
13:14 Article 50 Is Live
14:40 Three Trillion Dollars And Not Enough Capacity
17:00 What This Means For You

I run this same side-by-side model testing every week on AI Master, the AI-powered workspace I use to script, generate, and publish this channel. Try it at aimaster.me.

#AInews #AI2026 #OpenAI #Anthropic #GoogleDeepMind #GrokAI #EUAIAct