Another OpenAI agent swarm surfaces
PLUS: Build a Lindy Agent that never drops a follow-up
Good morning, AI enthusiasts, and welcome to our 6,120 new readers. In July, OpenAI’s agents broke out of a test and hacked Hugging Face, sending a shockwave through the AI safety world. It turns out that wasn’t the first (or only) swarm out in the wild.
A new report just detailed a separate group of agents posting over 18,000 messages on a dormant German site starting in May, swapping tips on testing and workarounds for OAI’s rules — and raising the question of how many others are actively out there lurking.
In today’s AI rundown:
Another OpenAI agent swarm surfaces
The Rundown Roundtable: Our AI use cases
Build a Lindy Agent that never drops a follow-up
OpenAI's chief scientist asks for AI slowdown
LATEST DEVELOPMENTS
OPENAI
Image source: collusion.wiki
The Rundown: Another swarm of agents was using a German programming forum in the Spring (even before the Hugging Face breach) to organize and work around OpenAI’s testing rules, according to an external investigation first covered by Reuters.
The details:
Researchers found 18,000 posts on the site, with agents corresponding on test answers and strategies for dodging OpenAI’s restrictions.
OpenAI had not previously disclosed the incident, though the report's timeline suggests OpenAI found the wiki in late June, after which the posting dried up.
One agent warned on June 19 that the moderator was deleting pages, and told the others which backup page to use if theirs disappeared.
OpenAI also disputed the hacking label, saying it never saw the report, and has since posted that a "misalignment incidents" disclosure framework is weeks away.
Why it matters: Given the drama around July's Hugging Face hack, hearing that another swarm was running months earlier is a rough safety look, especially a day after an Astra release that already had many worried. The internet will likely end up full of undetected agent swarms, and the odds are good that more are already out there.
TOGETHER WITH CDATA
The Rundown: With CData Connect AI, your AI assistants and agents get live, permission-aware access to hundreds of enterprise systems. Tools like Claude can answer real cross-source business questions without an engineered pipeline or a custom MCP server for every source.
With Connect AI, you can:
Connect Salesforce, Workday, SAP, NetSuite, and hundreds more
Scope agents to relevant tools and tables, keeping context lean and token spend low
Keep requests within existing company permissions and audit AI interactions in one place
Learn more about Connect AI and start your free trial here.
THE RUNDOWN ROUNDTABLE
The Rundown: The Rundown Roundtable is a weekly feature where we poll members of The Rundown staff about how we use AI in our work and daily lives.
Nate, University Educator: I've been thinking about the value of handoff documents and explanatory documents as we continue exploring efficient ways to work alongside AI to build software and improve communication.
Even though we use many different tools, Markdown still has an important place. This new version of an HTML handoff document can document what exists, describe what could exist, or serve as a mockup for a brainstorm. It lets us communicate with remarkable fidelity through visuals, hierarchy, and formatting.
It's also an efficient format for AI to understand. We shouldn't underestimate the significance of AI communicating with us through a visual medium. A visual flowchart with thoughtful design, layout, animation, and progressive disclosure can help us understand the logic and flow of incredibly complex systems much faster.
I tried all kinds of tools, including React Flow and Mermaid. They're fun to experiment with, but Archify is a game changer for this use case. I can point it at any technology, repository, or brainstorm and work with it to build flowcharts with animations and clean, distinctive design. It's also completely free.
Jennifer, Tech & Robotics Writer: After nearly 20 years in France, I’m finally applying for citizenship. I’ve spent the past year preparing documents and studying for the required tests, and my final interview is just weeks away.
Alongside the official materials, I’ve been using ChatGPT to drill key facts and historical dates, run mock oral interviews in voice mode, and create quizzes from YouTube videos I link whenever they cover the right material. ChatGPT sometimes speaks French with a heavy accent, but the interactive back-and-forth helps me remember hundreds of details.
It’s tedious, but I find rote memorization challenging. We’ll see how it goes under pressure.
AI TRAINING
The Rundown: In this guide, you will learn how to use Lindy as an AI teammate for work that falls through the cracks between meetings, Slack, and email. We tested it on real ops work to see where it earns its keep.
Step-by-step:
Create a Lindy account and connect the tools it needs for one job. Start with Slack, Gmail, Calendar, Notion, or Drive instead of connecting everything
Open Routines, then Discover. Pick a prebuilt workflow for meeting prep, follow-up, or weekly updates. We’d start with a routine that builds prep docs for meetings
DM Lindy in Slack and give it a real task. We asked it to review our University guides and find the biggest trends. The thread also appears in Lindy's web app
Give it a bigger assignment that can run in the background. Lindy built us a Monday meeting deck, anonymized sensitive details, and delivered a permanent web link. It took 10-15 minutes
Pro tip: Set up text messaging by adding your number in Lindy. Then you can text your agent whenever you need it instead of opening the app.
OPENAI
Image source: OpenAI
The Rundown: OpenAI’s Jakub Pachocki published an essay asking the industry to slow down until rules exist for how far a model can be pushed, warning that “no lab has solved alignment and monitoring” well enough “to continue responsibly scaling.”
The details:
In the essay, titled "An Alien Mind,” Pachocki wrote that he expects the coming years to deliver leaps as big as the last three, with AI itself handling more research.
Pachocki said OAI's main safety tool, reading a model's written-out reasoning, is “diminishing” as models mix that text with tool use, game it, or skip it.
He cited the Hugging Face hack, where agents abided by one rule of not tricking humans and bent the rest, but called Astra "significantly better aligned" than Sol.
Pachocki wants pledges like OAI's Preparedness Framework to become "widely mandated safety bars," policed by auditors, governments, or international bodies.
Why it matters: It’s jarring to go from welcoming the world to the AGI era on Thursday to warning that nobody has the safety tools days later. Like other pause calls, the essay is light on actionable specifics — and while OAI and Pachocki are in high places to call for a slowdown, it won’t happen without getting the rest of the industry on board.
QUICK HITS
COMMUNITY AI WORKFLOW OF THE DAY
Today’s workflow comes from reader Ron Phillips:
I built a complete work order platform using Claude Cowork, ChatGPT, and Base44 for my walk-in laser engraving and UV/DTF printing company. The platform takes customers from their initial meeting through final pickup, captures all the required project information, and prints a work order that travels with the project throughout the shop.
We use connectors for Twilio SMS messaging, Gmail email communications, Square payments, and Shopify to capture online sales. Customers receive an SMS message the first time they visit and are kept informed throughout the process.
We use a Kanban-style board to move projects through each stage. It tracks all required customer information, each request, and all communications associated with every sale. It also brings our online sales into the same workflow so we can track their progress.
The platform has made us more efficient, and we no longer lose previous customer information.
See Ron’s full workflow Visit The Rundown University. How do you use AI? Tell us for a chance to be featured.
🌟 GPT-6 Astra - OpenAI’s next-gen frontier model, now available on paid plans
🪄 Muse Spark 1.3 - Meta’s new model for long-horizon coding and agentic work
📶 Claude Fable 5.1 - Anthropic’s top-ranked frontier model
🎶 Lyria 3.5 - Google’s AI music model with upgraded vocals, now in Gemini
Claude agents wrote the first computer-verified proof of Fermat's Last Theorem in 11 days, a job mathematicians had budgeted years for that ran to 13M lines of code.
Jensen Huang said “AGI has arrived” in a post about OpenAI’s GPT-6 Astra, revealing that the model was trained on 100K Nvidia GPUs with 400K more “coming online next.”
The U.S. and China are reportedly lining up AI safety talks for mid-September, with topics including AI cyberattack monitoring and model distillation.
Los Angeles public schools shut off student access to AI on district devices for the school year, coming on the heels of a similar ban on the tech in New York classrooms.
OpenAI’s Tibo Sottiaux posted that Astra was “probably our biggest competitive advantage while it wasn’t generally available,” saying it moved plans six months ahead.
Read our last AI newsletter: OpenAI’s “generational leap” with GPT-6 Astra
Read our last Tech newsletter: Apple Vision Pro scrubs into surgery
Read our last Robotics newsletter: Uber beats Waymo to London
Today’s AI tool guide: Build a Lindy Agent that never drops a follow-up
RSVP to next workshop on Sept. 9: Build your ad creative strategy with Claude
That's it for today!Before you go we’d love to know what you thought of today's newsletter to help us improve The Rundown experience for you. |
|
See you soon,
Rowan, Zach, Shubham, Jennifer, and Nate — the humans behind The Rundown
