Leaked Chats, Watching Lampposts
· The Fluency Briefing
The Fluency Briefing
Your Guide to What's Happening in AI and Why It Matters to You
Tuesday, July 28, 2026
You shared a Claude chat with a coworker last month, right?
Turns out Google could read it too - along with hundreds of other people's private conversations. Meanwhile, AI-powered lamp-posts in Britain are learning to recognize your face, and Google wants to plan your dinner party. Tuesday's theme: AI is now deeply, uncomfortably personal.
Today in AI:
- Your Claude Chats Were Googleable - Redditors discovered that hundreds of shared Claude conversations were indexed by Google, exposing private legal questions, crypto wallet keys, and deeply personal exchanges to anyone who searched. Anthropic is scrambling to respond. Apple
- Britain's Lamp-Posts Are Watching - A UK company called Conflow Power Group is rolling out solar-powered smart streetlights with facial recognition, number plate reading, and potential gait analysis. The home secretary has already endorsed a "panopticon state." theguardian.com
- Google Wants to Host Your Dinner Party - Google's AI Mode in Search now generates tablescapes, menus, and shopping lists for dinner parties, complete with visual previews via its Nano Banana image tool. Dinner party searches are trending hard in 2026. blog.google
- Everyone Wants to Touch Grass - Google Trends data shows searches for "how to digital detox" are up 110% this year, and "run club" hit all-time search highs. The irony: people are using AI-powered search to figure out how to stop using screens. blog.google
- AI Still Can't Make You Laugh - A new benchmark called LaughBench tested frontier models on original joke generation. None produced genuine laughter. The researcher argues humor requires modeling another mind's false inferences - a task current AI fundamentally can't do. Curalinc
- Cursor Bets Big on India at $7/Month - AI coding tool Cursor launched a localized India subscription at roughly one-third its standard price, ahead of its SpaceX acquisition. India is already Cursor's third-largest market, with its user base tripling in the past year. techcrunch.com
- The NYT's $20M AI Copyright Fight Grinds On - Publisher A.G. Sulzberger told Wired the Times' lawsuit against OpenAI and Microsoft is about defending journalism broadly, not just the Times. The paper now has 13 million subscribers and is still filing discovery motions nearly three years in. wired.com
- AI Rollup Startup Dwelly Raises $170M - The UK proptech company acquires traditional letting agencies and layers AI over their operations, now managing 15,000 properties. Angel investors include founders from ElevenLabs and Synthesia. sifted.eu
Today's Takeaway:
Anthropic's shared-chat indexing exposed conversations that users reasonably believed were private - including a lawyer contemplating self-reporting a conduct breach, per Apple. The same week, Conflow Power Group's smart lamp-posts are shipping with facial recognition and gait analysis to UK streets, as theguardian.com reports.
These aren't the same failure, but they share a root cause: AI tools that collect or expose personal data are being deployed faster than the consent architecture around them. Your chatbot assumed a share link was access-controlled. Your streetlight assumed a home secretary's endorsement was sufficient public debate. Neither asked you first.
The uncomfortable bet for anyone building on these platforms: the default privacy posture of most AI products is "open until someone complains," and the complaints are now arriving with screenshots attached.
💡 Fluency Moment - Building your AI fluency, one term at a time.
"Benchmark"
In plain English: A standardized test used to measure and compare how well AI systems perform. Think of it like: Like a driving test - every candidate faces the same challenges so scores mean something fair. Why you'll hear about it: AI companies use benchmarks to claim superiority; knowing this helps you spot marketing spin.
The Bottom Line
The Pattern: We've tracked the gap between AI's capabilities and its human guardrails for weeks. What's new is that the exposure isn't theoretical anymore - it's your actual chat logs sitting in Google's index and your face being processed by a streetlight. The infrastructure is live before the consent conversations have started.
The Other Read: The Claude indexing bug is a straightforward engineering oversight with shared links, not evidence of systemic disregard for privacy - Anthropic will patch it, Google will de-index the pages, and most users never share chats at all. Fair point, but the lawyer whose conduct-breach question was publicly searchable won't find much comfort in a hotfix. We lean toward this being a symptom, not an outlier.
Your Move: Go to claude.ai right now, click your profile settings, and check whether you have any shared conversations active - takes 90 seconds. Revoke any you don't need public. If you shared something sensitive, search "site:claude.ai/share" plus a keyword from that chat to see if Google cached it.
What We're Working On
✨ Founding Cohort Special - 60% Off! - Use code MAF20 to join for just $20/month (regularly $50). Get weekly group sessions & workshops, self-paced courses for all levels, access to tools & templates, challenges with peer feedback, and 24/7 support community. → Join Now
✨ Free 30-Minute AI Consultation - Discover how My AI Fluency can help your business unlock the potential of AI. We'll discuss your goals, explore practical AI opportunities for your industry, and outline clear next steps. → Schedule Free Call
✨ How AI-Fluent Are You? - Test your AI fluency with our interactive quiz. See how you stack up and discover what to learn next. → Take the Quiz
💬 Community | 📞 Book a Consultation | 🌐 Website
Fluently yours, The My AI Fluency Team