Newsletter Content Required
· The Fluency Briefing
Welcome back to your essential weekly roundup of
📰 The Big Story
Here's the thing about building something too powerful to release: you still have to tell the world about it.
Anthropic announced Project Glasswing this week, a sweeping cybersecurity initiative built around an unreleased frontier model called Claude Mythos Preview venturebeat.com, Apr 8. This model is reportedly so proficient at identifying and exploiting software vulnerabilities that Anthropic decided it was too dangerous for public access. Instead of shelving it quietly, Anthropic turned it into a coalition project, partnering with Apple, Google, and more than 45 other organizations to use the model as a shared stress-test for AI cybersecurity defenses wired.com, Apr 8.
Translation: Anthropic built a digital skeleton key, then invited the neighborhood over to help them figure out which locks it could pick.
Now contrast that with what happened at ProPublica. Unionized staff at one of America's most respected investigative newsrooms walked off the job for 24 hours, demanding protections around AI use, layoff safeguards, and better wages theverge.com, Apr 8. These aren't Luddites smashing looms—they're journalists watching AI capabilities sprint ahead while workplace policies limp behind.
Put these two stories side by side and the picture gets uncomfortable. On one end, an AI lab acknowledges its own creation is too dangerous for the public. On the other, workers at a news organization are begging for basic guardrails that don't exist yet. The technical capability is here. The institutional frameworks to manage it? Still loading. That gap—between what AI can do and what society is ready for it to do—defined this entire week.
📋 5 Stories That Shaped the Week
Beyond the headlines, here's what shaped the week...
Google made a power move by fully baking NotebookLM into the Gemini app, turning its AI chatbot into a persistent research and project management hub with a new "notebooks" feature engadget.com, Apr 9. The Verge confirmed that notebooks let you pull in conversations, sources, and files into organized workspaces that Gemini remembers across sessions theverge.com, Apr 9. The "so what" here is significant: Google is no longer positioning Gemini as a chatbot. It's positioning it as your second brain—one that never forgets a thread.
While Google was deepening integration, Microsoft was quietly asserting independence from OpenAI. The company launched three entirely in-house AI models—MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2—covering speech transcription, voice generation, and image creation venturebeat.com, Apr 4. Let's be real: this is Microsoft hedging its $13 billion OpenAI bet by building its own foundation. The message to partners and competitors alike? We don't need anyone's permission to compete.
Meanwhile, two former Apple Vision Pro engineers debuted "Button," a $179 wearable AI puck that looks like an iPod Shuffle and only listens when you physically press it wired.com, Apr 9. In a market littered with the corpses of always-listening AI gadgets (rest in peace, Humane Pin), Button's bet on consent-based interaction is a direct rebuke to ambient surveillance. The real question: will consumers pay for privacy, or has that ship sailed?
On the investment front, the AI gold rush is dragging private wealth into increasingly risky territory, with family offices and high-net-worth individuals pouring capital into earlier-stage AI bets than ever before techcrunch.com, Apr 7. And the geopolitical stakes keep rising—a BBC analysis mapped how the US and China are each winning different AI races, with either nation capable of pulling ahead depending on which capability matters most in the next 18 months bbc.com, Apr 7. The takeaway: the AI competition isn't one race. It's a decathlon, and nobody's winning every event.
🔗 The Pattern We Noticed
Connecting the dots...
The thread running through this week? The gap between capability and control. Anthropic built an AI hacker it won't release. ProPublica's staff walked out because workplace AI policies don't exist yet. Microsoft built its own models to reduce dependency on a partner it can't fully control. Button built a wearable specifically designed to give you control over when AI listens.
Why now? Because AI capabilities are compounding faster than any single institution—corporate, governmental, or labor—can write rules for them. Every story this week is a different actor trying to grab the steering wheel of a car that's already moving.
For you, this means the competitive advantage isn't adopting AI first. It's adopting it with guardrails that actually hold. The businesses building internal AI policies today—around data use, workforce impact, and security—will be the ones standing when the regulatory dust settles. The ones winging it? They're one "too-dangerous-to-release" model away from a very bad quarter.
🔮 On the Horizon
These stories are still unfolding — here's what to track:
- Anthropic's Project Glasswing coalition: Watch whether the 45+ partners actually share vulnerability data openly or treat it as competitive intelligence. The first real test reports should surface within weeks.
- ProPublica contract negotiations: The strike was 24 hours, but the demands—AI protections, IP safeguards—will set precedent for every newsroom negotiation that follows theverge.com, Apr 8.
- Google Gemini notebooks adoption: If persistent memory workspaces gain traction, expect Microsoft and OpenAI to fast-follow with their own project-layer features by Q3 theverge.com, Apr 9.
📚 Term of the Week
Going deeper on one concept that shaped this week's AI conversation.
"Red Teaming"
What it is: Red teaming is the practice of deliberately attacking a system—whether software, a network, or an AI model—to find vulnerabilities before bad actors do. In AI, this means having humans or other AI models probe a model's outputs for dangerous capabilities, biases, or exploitable weaknesses. Think of it as hiring a professional burglar to test your home security.
Why it matters this week: Anthropic's entire Project Glasswing is essentially red teaming at industrial scale—using Claude Mythos Preview to stress-test cybersecurity across 45+ organizations wired.com, Apr 8.
The bigger picture: As AI models grow more capable, red teaming is shifting from a nice-to-have to a regulatory expectation. Expect it to become a standard compliance requirement, not just an internal best practice.
Try this: Ask your AI assistant to "find three weaknesses in this business plan" and see how critically it can evaluate your own work.
📬 That's a Wrap
That's a wrap on this week, and the uncomfortable truth it delivered: AI is now building things its own creators won't release, while the rest of us are still arguing about the basics.
Your move: Draft a one-page internal AI use policy for your team this week. Cover three things: what data you'll share with AI tools, what outputs require human review, and who's accountable when something goes wrong. Start messy. Refine later. Just start.
Fluently yours, The My AI Fluency Team
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
💬 Community | 📞 Book a Consultation | 🌐 Website
