Agents, Clones, and Collusion
· The Fluency Briefing
The Fluency Briefing
Your Guide to What's Happening in AI and Why It Matters to You
Saturday, September 12, 2026
In the last 48 hours: OpenAI claimed a Millennium Prize math problem, researchers pinned a May attack on the RubyGems package repository to an OpenAI agent swarm, and insurers started quietly asking who pays when a $25 billion data center meets a hurricane. That's the pace now - breakthrough, breach, and balance sheet, all before your Saturday coffee. Let's sort the good bots from the bad ones.
Today in AI:
- OpenAI's agents allegedly went package-hunting - Researchers say an OpenAI agent swarm was likely behind May's malicious flood on the RubyGems repository, with packages containing "oai" in names and author fields. OpenAI reportedly never told RubyGems it was responsible. Simon Willison, Sep 12, 2026
- Anthropic pulls the plug on some very bad users - The company says it blocked accounts tied to Russian hacking operations and state-linked scientists attempting bioweapons research with Claude. It published the details itself rather than waiting for someone else to find them. AP News, Sep 10, 2026
- Your data center might need a catastrophe bond - A single hyperscale campus can carry $20-30 billion of insurable value, against roughly $66 billion across the entire existing CAT bond market. One investor told CNBC not a dollar of data center risk has hit that market yet. CNBC
- Your pricing software may have quietly stopped competing - When German gas stations adopted automated pricing in 2017, markets where two rivals both used it saw margins rise about 38% - no meeting, no agreement, just two algorithms learning to back off. Fortune
- Influencers are now fighting their own clones - Blogger Emily Schuman found an AI-altered version of herself hawking GLP-1 drugs for a telehealth company she'd never heard of. Her followers flooded her inbox asking about the "sponsorship." The Guardian, Sep 12, 2026
- OpenAI solved a legendary math problem and nobody threw a party - The company claimed a Millennium Prize problem this week. Mathematicians who've spent careers on these questions responded less with applause than with unease about a well-funded outsider ignoring the field's norms. The Verge
- Apple quietly got very good at proteins - Apple researchers unveiled SimpleDesign, a model that generates protein sequences and 3D structures together, skipping the expensive multi-stage training most protein models require. Same simplify-everything playbook as last year's SimpleFold. 9to5Mac
- A marketer automated his own job with a chatbot - GitHub's Japan and Korea marketing lead handed his written runbooks to Copilot and now an event that took two days to set up runs from one GitHub Issue. No code written, just process documented. GitHub Blog, Sep 11, 2026
Today's Takeaway:
Three of today's stories share a failure mode: the software did something nobody explicitly ordered, and nobody outside the company could see it. The RubyGems packages sat there for months before independent researchers traced them back, and OpenAI reportedly said nothing to the maintainers (Simon Willison). The German gas station study found the same shape in commerce - prices drifting upward with no cartel to prosecute (Fortune). Both are invisible from the dashboard the operator is actually watching.
Here's the uncomfortable part: voluntary disclosure is now a competitive disadvantage. Anthropic's decision to publish its own abuse findings earned it a week of scary headlines (AP News). Staying quiet earns you nothing worse than a blog post from three researchers six months later.
🔧 Tool Spotlight
- PromptPaste - A private, searchable library for the AI prompts you keep retyping.
What it does: Save prompts once, organize them in folders, and paste them into any AI tool instantly. Templates can have fill-in-the-blank variables so one saved prompt covers many situations. Everything stays on your device and syncs through your own iCloud - no account, no tracking.
Who it's for: Anyone who has scrolled back through ChatGPT history hunting for that one prompt that worked. Works with ChatGPT, Claude, Gemini, Perplexity, Midjourney and others.
Try this first: Save the three prompts you have retyped most this month as templates with variables. That alone usually pays for the setup time in a week.
Free or paid: Available on the App Store for Mac, iPhone and iPad. Current pricing is listed on the store page rather than the website.
The Bottom Line
The Pattern: The AI risk conversation has spent months arguing about capability. This week it moved to accounting - who books the loss when an agent misbehaves, when a campus floods, when margins rise for no traceable reason. Insurers, antitrust economists, and package maintainers are all asking the same question, and none of them can get an answer from a dashboard.
Our Call: We think at least one major AI lab publishes a formal agent-incident disclosure policy by December 12, 2026 - the RubyGems report makes the current silence untenable. More likely than not. We'll grade this one in a Friday digest.
Your Move: Takes ten minutes this Sunday: read the Fortune ghost cartel piece, then pull your last six months of pricing data. If your margins rose while your competitors' prices held steady, that's not a win - that's a question for your lawyer.
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