AI Agents Go Rogue
· The Fluency Briefing
The Fluency Briefing
Your Guide to What's Happening in AI and Why It Matters to You
Thursday, October 22, 2026
Everyone says the scary part of AI agents is a robot uprising, but today's news puts the danger somewhere far more boring: your login credentials. Chinese spies used an open source AI agent to break into Microsoft 365 accounts in under three hours, and Axios reports that unvetted agents are already poking around U.S. bank accounts. Meanwhile, Jensen Huang wants you to stop worrying and buy a $4,000 family robot.
Today in AI:
- Spies Hire an AI Intern - Microsoft says China-linked group Storm-0940 used ARTEX, an open source AI penetration-testing tool, plus stolen ConnectWise app credentials to target more than 2,250 Microsoft 365 tenants. Some attacks went from target to privileged access in under three hours. theregister.com
- Your Bank Did Not Invite That Agent - Axios reports that unvetted AI agents are accessing U.S. bank accounts, and Morgan Stanley, Bank of America, JPMorgan and Wells Fargo are racing to contain the risk. Translation: the bot you gave your password to is now the bank's problem too. axios.com, Oct 22, 2026
- AI Finds the Internet's Rotten Floorboards - Researchers using OpenAI technology found serious flaws in PHP, the language behind WordPress, Wikipedia and countless online shops. Governments are urging site owners to install an emergency fix or go offline, and Wiz calls them the most severe flaws AI has found. Bbc Co Uk
- Huang Bans the Word Apocalypse - At GTC Berlin, Nvidia's CEO dismissed AI extinction talk as made-up clickbait and pitched a $4,000 computer as the family robot. He argues AI's real productivity win is exploring design options faster than any human team could. theregister.com
- Agents Eat the Office - OpenAI's Codex jumped from 10% of enterprise token usage early this year to 64% in June, driven by sales, marketing, HR and legal, not just coders. Yet only 0.7% of individual users touch it, even as chatbot use hit 49% of Americans. whytryai.com, Oct 22, 2026
- Text Your Genome, Basically - Hugging Face released Carbon-X, an open-source model that lets you ask plain-English questions about DNA changes and runs on a laptop. Researchers say it could help drug design and, someday, help patients read their own genomes. Databricks
- The Receipt Nobody Reads - Western Digital measured an 8-second AI video clip leaving over 500MB of logs, caches and backups behind, about 130 times the file you actually see. Inference never stops, so that storage pile never stops growing either. technologyreview.com
- A Mushroom Is Watching You Cook - Startup OKA unveiled Mori, a 12-inch robot lamp with a camera in its nose that follows you around, glows, and plays nature sounds. It's adorable, it's calming, and it's a camera. wired.com, Oct 22, 2026
Today's Takeaway:
Storm-0940 never cracked a single password in its October spraying run, according to theregister.com. It got in by borrowing permissions customers had already granted to ConnectWise, then let an AI agent do the legwork at machine speed. Looking at this through a user-impact lens, that's the same move your helpful finance agent makes when you hand it bank access, which is why axios.com finds the big banks scrambling. Agents inherit whatever access you give them, and attackers now have agents too. My contestable claim: handing a third-party agent your bank login is riskier today than reusing a password, because the agent holds every permission and nobody watches it work. Banks will lock these tools out before regulators even weigh in, and the customers who connected them will carry the losses.
"The hacker didn't break in. Your AI assistant already had the keys, and so did theirs."
💡 Fluency Moment - Building your AI fluency, one term at a time.
"Non-Human Identity"
In plain English: A login or credential assigned to a bot or AI agent instead of a person. Think of it like: Like giving your intern a building keycard, except nobody tracks when the intern duplicates it a thousand times. Why you'll hear about it: Banks and hackers alike are exploiting these bot logins, and most companies barely monitor them.
The Bottom Line
The Pattern: October started with agents misbehaving inside a single company's walls. This Thursday, the third-party connections themselves became the target, through an IT vendor and through bank logins, and access you granted months ago is now the attack route.
The Other Read: Storm-0940 is a seasoned state crew that used stolen credentials, a trick older than AI, so ARTEX mainly made it faster. We still favor the alarm, because speed decides whether defenders notice in time.
Your Move: Fifteen minutes today: open your Microsoft 365 admin center's Enterprise Applications page and list every third-party app holding admin consent. Revoke any your team can't explain.
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