Autonomous Agents, Courtroom Evidence
· The Fluency Briefing
The Fluency Briefing
Your Guide to What's Happening in AI and Why It Matters to You
Saturday, September 26, 2026
By this time next year, you'll be talking to glasses that can't see you, delegating whole afternoons to a model that works unsupervised, and explaining to someone why your AI-generated song is not evidence of anything. Two of those futures shipped this week. The third already happened in a courtroom.
Today in AI:
- Anthropic's new Opus works the night shift - Claude Opus 5.5 landed with a pitch of top-tier performance at 40% lower cost than Opus 5, plus what Anthropic calls extremely strong cyber capabilities. Reviewers highlight its ability to run on its own for extended stretches without babysitting. lesswrong.com, Sep 26, 2026
- Enterprises can see their agents. Stopping them is harder. - Companies have gotten decent at monitoring what AI agents do, but far less capable of interrupting one mid-task. A bipartisan bill from Rep. Gottheimer would push for human-in-the-loop controls on rogue agents. Hypr
- Meta's answer to "pervert glasses": remove the camera - At Meta Connect, the company demoed unreleased audio-only smart glasses with six microphones and no camera, after criticism that its camera specs invited surveillance. They also hook into Muse, Meta's agent, via voice commands. Meta
- The Pope has notes on your roadmap - Pope Leo XIV warned during his France trip against building a "paradise of machines" that sidelines human dignity. It's the loudest moral-authority intervention on AI since his election, and it landed alongside talks with Macron. apnews.com, Sep 25, 2026
- An AI love song got entered into evidence - Prosecutors at Caleb Flynn's murder trial played two AI-generated songs he made for his mistress, recovered from his phone with forensic tool GrayKey. A forensic investigator testified about the keys. The courtroom sat in silence afterward. 404media.co, Sep 25, 2026
- Your twenty years of location history just became ad-tech - Foursquare and Swarm are being acquired by Infillion, an ad-tech firm, prompting longtime users to bail. One writer noted he's been backing up check-ins to spreadsheets for a decade, and now Gemini is his memory. spyglass.org
- China corners the robot hand market - Chinese suppliers have taken the lead in dexterous hands, the fiddly multi-jointed components that let humanoid robots actually grip things. It's the part of the robot stack where manufacturing scale beats clever software. Gongboshi-robot
- Muse wants you to talk like a drill sergeant - TechCrunch's reporter found Meta's agent needs very direct verbal commands to cooperate, and Meta plans to let you pick a cute avatar to represent it. Charming, slightly comical, and not yet public. Meta
Today's Takeaway:
Rep. Gottheimer's bill is specifically about stopping agents, not watching them - which tells you which half of the problem industry already solved. Monitoring is a dashboard. Interruption is an architecture decision, and most enterprise deployments never made it (Hypr).
That gap gets more expensive as models get cheaper to leave running; the whole selling point of Anthropic's latest Opus is persistence without supervision (lesswrong.com). Here's the part vendors won't say: an off switch buried three menus deep in an admin console is not an off switch, it's a compliance screenshot.
If your team can't name the person who can kill a running agent in under sixty seconds, you don't have controls - you have logs.
💡 Fluency Moment - Building your AI fluency, one term at a time.
"Human-in-the-Loop"
In plain English: A required checkpoint where a person must approve or stop an AI's action before it proceeds. Think of it like: Like a co-pilot who must confirm before autopilot executes a risky maneuver mid-flight. Why you'll hear about it: Lawmakers want it mandatory as agents like Opus 5.5 run longer unsupervised.
The Bottom Line
The Pattern: Third week running, the binding question isn't what AI can do but who holds the brake - and this time a US congressman and the Pope arrived at the same worry within 48 hours, from opposite ends of the authority spectrum. Moral pressure and legislative pressure converging is new; neither one writes your vendor contract.
Our Call: No major agent platform will ship a documented one-click kill switch with a published stop-latency number before December 31, 2026. More likely than not. We'll grade this one in a Friday digest.
Your Move: Fifteen minutes this Sunday: open the admin console of whichever AI agent tool your team uses most, find the stop or revoke control, and time how many clicks it takes from login. Send that number to your ops lead.
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