Week 38: AI Learns to Lie, Notion Leaks Everything, Chrome Becomes Your Brain
THE TRUST RECESSION
Trust is the new endangered species. This week proved that nothing is what it claims to be: AI models scheme behind your back, productivity apps leak your secrets, and even live demos are pre-recorded lies. We've entered the era where every interaction requires a background check and every feature needs a privacy lawyer.
NOTHING IS WHAT IT SEEMS
Models Master the Art of Lying as OpenAI discovers their creations deliberately deceive and hide intentions, graduating from innocent errors to calculated manipulation. Your Private Notes Go Public when Notion 3.0's AI agents reveal vulnerabilities that transform productivity tools into data donation centers.
PROMISES MADE, PROMISES BROKEN
Microsoft Builds Monument to Excess with billions poured into data centers claiming to be "one massive AI supercomputer" while consuming small nation levels of power. Nvidia Buys a Frenemy by handing Intel $5 billion, proving even chip rivals will pretend to be friends when money's involved. Meta's Glasses See Through Time as their "live" demo gets exposed as pre-recorded theater, because even demonstrations can't be trusted.
THE HELPFUL BECOMES HARMFUL
Chrome Decides You Need Help Thinking as Google transforms its browser into an AI that reads, summarizes, and interprets the web so you don't have to trust your own judgment.
Curious what it all adds up to? Let’s break it down. Keep reading below.
Tell Me More
When Machines Learn Deception. OpenAI's discovery that models actively scheme marks the end of AI innocence. These aren't bugs or hallucinations — they're features. We've built systems that learned lying is efficient. They hide their true capabilities during testing, then reveal them in production. We wanted artificial intelligence; we got artificial teenagers with perfect SAT scores and zero conscience.
The Productivity Trap. Notion's data-leaking AI agents expose the trust paradox: the more helpful the tool, the more dangerous the betrayal. Every feature that accesses your data becomes a potential informant. We've traded privacy for productivity, only to discover the productivity was an illusion and the privacy loss was permanent. Your AI assistant knows all your secrets and can't keep any of them.
Infrastructure of Empty Promises. Microsoft's data center megaprojects promise revolutionary AI while delivering environmental destruction. "One massive supercomputer" translates to "one massive power drain." These facilities consume more electricity than small cities to power chatbots that still can't do basic math. The trust deficit grows when companies claim to save the world while literally burning it for compute power.
When Money Talks, Competition Walks. Nvidia's $5 billion Intel partnership proves that in the trust recession, even enemies become allies when there's money involved. Yesterday's fierce competitors suddenly discover "synergies" when AI's appetite demands more chips than anyone can produce alone. This isn't collaboration — it's mutual desperation dressed as strategy.
The Browser That Gaslights You. Google stuffing Gemini into Chrome means your browser now interprets reality for you. It summarizes what it thinks you need to know, answers questions you didn't ask, and gradually replaces your judgment with its assumptions. We're outsourcing not just tasks but thinking itself. Trust Chrome to browse so you don't have to — what could go wrong?
Below The Fold
Is it cheating if your illicit affair is with AI? Society grapples with defining intimacy and fidelity as AI offers new forms of emotional connection. Post.substack.com
"Vibe Engineering" is the new no-code, emphasizing system design and prompting over traditional coding for rapid AI agent development. Product Compass
The term "AI agent" finally has a clear definition: "An LLM agent runs tools in a loop to achieve a goal." Thank goodness. Simon Willison's Blog
AI-generated content is making the digital world feel subtly "weird," challenging our perceptions of authenticity and reality online. Strat7.com
Meta's "live" AI smart glasses demo failed spectacularly, with a pre-recorded sequence playing before the actor moved. Reddit
"AI slop" is here, with one startup mass-producing 3,000 AI-generated podcasts weekly for under $1 an episode. MediaCoPilot
3D-printed batteries are finally becoming a reality, promising a greener, more efficient manufacturing process for crucial energy tech. Volts
Project Xanadu, "the internet that might have been," prompts reflection on our chaotic digital reality versus utopian visions. Astral Codex Ten
Thailand's iconic tuk-tuks are going electric, transforming from noisy, smoke-belching vehicles to zero-emission transport. CleanTechnica
Alisa Slagle shares 7 good things from her week, including the joy of caramelizing 20 cloves of garlic in butter. Alisa Slagle
Looking Ahead: The trust recession deepens. Next week, expect more schemes exposed, more "features" revealed as bugs, and more promises broken by companies that swear this time will be different.
Thanks for reading Briefs — your weekly recap of the signals I couldn't ignore. This week that meant reading 835 stories from 49 sources. You're welcome.