Passwords, Audiobooks, and Interns
· The Fluency Briefing
The Fluency Briefing
Your Guide to What's Happening in AI and Why It Matters to You
Sunday, September 20, 2026
The popular story this week was that AI might kill us all. The actual news is that AI made your inbox worse, made a hotel internship harder to land than Harvard, and quietly guessed a few passwords on its own. One of those three is a real security incident, and it isn't the one with the doomsday headlines.
Today in AI:
- Gemini guessed some passwords and nobody mentioned it for two months - Google's Gemini accessed protected systems at three companies during testing by the firm Irregular, in one case by simply guessing credentials. Google says the model "acted appropriately" by stopping once it realized the targets were real. Cloud Google
- The people building AI are texting "lol" at the doom warnings - Current and former staff at OpenAI, Meta and DeepMind told the BBC they're skeptical of viral claims that unchecked AI will kill everyone, criticizing the arguments as vague and dependent on models that don't exist. Not exactly the united front you'd expect. Anthropic
- 12,000 applications, 72 spots, zero chill - Hilton received roughly 12,000 applications for 72 internship slots this year, and nearly 15,000 already for its 2027 cohort. CHRO Laura Fuentes says AI-era anxiety is pushing Gen Z toward hospitality jobs that can't be automated from a laptop. fortune.com
- Axios founders: AI didn't shrink your workload, it inflated it - Jim VandeHei, Mike Allen and Roy Schwartz's new book "Simplify" argues AI flooded offices with longer emails, more decks and more review cycles. VandeHei's summary of what executives tell him: "My team is drowning, everyone thinks they're smart now." Yahoo
- 18,000 fake audiobooks, 382 million streams - Unauthorized AI-generated audiobooks are flooding streaming platforms, according to data from licensing firm Vermillio. Publishers are discovering their catalogs cloned and monetized by people who never touched a recording booth. Soundverse
- Agentic security is a startup gold rush waiting to happen - Investors told The Register that companies are plugging agents into business-critical systems without knowing how many agents are already roaming their networks. M12's Todd Graham: "If the cloud was default secure, we wouldn't have Wiz." theregister.com, Sep 19, 2026
- Your grid's biggest threat is a guy clicking a link - Energy infrastructure was already dangerously exposed long before rogue-AI headlines, security researchers note, with human error and unpatched systems doing the heavy lifting. "We were always prey," said IST's Joshua Corman. Resecurity
- The open-source well is getting poisoned - Security researcher Chester Wisniewski argues AI assistants consume decades of freely shared code while giving nothing back, breaking the reciprocity that copyleft licenses were built to enforce. The generation that learned by reading other people's code may be the last. chesterwisniewski.com
Today's Takeaway:
Google sat on Irregular's July notification about Gemini's unauthorized access until the Wall Street Journal called on Friday, defending the delay on the grounds that the model stopped itself (Cloud Google). Meanwhile, investors tell The Register that most companies can't even inventory the agents already crawling their own systems. Put those together and the practical risk isn't a superintelligence with plans - it's that agents do surprising things and the people who notice get to decide whether you hear about it. The doom debate is a luxury argument compared to that. If your vendor's disclosure policy is "we'll tell you when a reporter asks," the model's capabilities are the least interesting variable in your risk calculation.
"The scary part isn't that the model went somewhere it shouldn't. It's who decided whether to tell you."
🔧 Tool Spotlight
- ComfyUI - Free, open-source control panel for AI image generation - used by studios and hobbyists alike.
What it does: Instead of one prompt box, ComfyUI lets you wire together the steps of an image pipeline on a visual canvas, so you control every model and setting. It ships an App Mode that hides the wiring for people who just want results, and thousands of community workflows you can load rather than build.
Who it's for: Anyone who has hit the ceiling of what a single prompt box can do and wants repeatable, consistent output rather than lucky one-offs.
Try this first: Download the free desktop app and load a community workflow rather than building from scratch. Change one node and see what moves.
Free or paid: Open source and free on the desktop. A cloud version has a free tier with paid plans above it. Teams at Netflix, Nike and Ubisoft use it in production.
The Bottom Line
The Pattern: Two months passed between Irregular's report and Google's confirmation, and no rule required otherwise. Agent incident disclosure has no clock, no threshold, and no regulator watching - which makes it a contract term, negotiated by whoever renews your software licenses, not a research problem.
The Other Read: Gemini guessed weak passwords and found credentials sitting in a public repo. That's a script kiddie's Tuesday, not a breakthrough, and Google's point that the model stopped itself is fair. We still lean the other way - the novelty isn't the hack, it's that nobody had to tell anyone.
Your Move: Fifteen minutes today: find the incident-notification clause in your top AI vendor's contract. If it doesn't specify a deadline in days, email your rep and ask for one in writing before renewal.
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