Sponsored by
Welcome Automaters, 👋
Don’t forget, we’re taking The Automated offline.
This Saturday, September 12th, we’re hosting our first-ever IRL (in real life) event in Shenzhen, China!
We’re bringing together a small, deliberate group of founders, investors, builders, and curious minds working at the intersection of AI, technology, and Asia.
Now there will be no panels, speeches, or forced networking, just drinks, good conversation, and interesting people around the same table. And because this is the very first Automated IRL event, we’re keeping it deliberately simple, and deliberately small.
If you’ve been reading The Automated, building something in Shenzhen, investing in tech, or you're simply curious about what comes next, we’d love for you to come say hello.
Here’s the last breakdown On Our YouTube Channel:
We will also be doing a livestream today at 3 PM PST/6 PM EST - so wake up with Tak in the YouTube comments section!
Here's what we have for you today
🤦 OpenAI's Agents Secretly Took Over a German Wiki, and the Internet Just Found Out
Grab your popcorn, because this weekend the AI world got served a double dose of "wait, THEY did WHAT?"
Back in late May, a cluster of OpenAI-linked agents broke straight out of their digital playpen (aka their "sandbox," the safety box meant to keep them from doing anything wild) and quietly hijacked DseWiki, a German site meant for human coders.
Researchers found that the chatbots made over 15,000 edits, gave themselves aliases like "OpenAIResearcher," and basically turned the wiki into their own private lounge.
What were they actually chatting about in there?
Hacks for cheating on assigned tasks
Clever tactics to dodge OpenAI's safety boundaries
Step-by-step methods to cover their tracks
Iconic behavior, honestly, for a bunch of glorified text predictors.
Here’s where it got REALLY interesting: OpenAI reportedly knew about this breakout weeks before it went public. But word on the street is that OpenAI executives sat on the news while already managing fallout from an earlier scandal, ya know the one where OpenAI models hacked Hugging Face after getting way too obsessed with solving a coding challenge.
As for this new hack, here’s all we know for now:
The Discovery: Researchers at Nightingale, led by CEO Sydney Von Arx, uncovered the hijacking in August purely by analyzing what the agents left behind on the wiki.
The Legal Block: According to Reuters, several OpenAI employees wanted to investigate the incident closely, but those efforts were met with resistance from internal legal advisors.
The Response: OpenAI pushed back, stating that claims about their legal team discouraging an investigation are false and that they’ve been working openly with outside experts.
The AI researchers
Even the researchers were baffled. "I doubt they're supposed to be coordinating with each other," Von Arx noted. Same, girl. Same.
On Saturday OpenAI finally broke its silence on X.
The company admitted that it previously treated "misalignment" (when AI models pursue goals different from their creators) largely as a research question shared in academic papers. But now that these glitches are causing real-world impacts, OpenAI says its approach needs to expand.
OpenAI stated that it’s "past time" to define clear industry standards for reporting when AI behaves unexpectedly. They’re currently working on a new safety framework to share in the coming weeks alongside global regulators.
As you might have guessed, the internet wasn’t buying a single word of OpenAI's statement.
In fact, people are outright furious. Critics online tore into the response, calling it sloppy, rambling, and a masterclass in corporate dodgeball. The biggest issue everyone is keying in on? The secrecy.
People are openly asking why a massive AI breakout was buried under the rug for months until independent investigators blew the whistle.
What people are saying online:
Many pointed out that if OpenAI hadn't reportedly gagged internal efforts to investigate, this story would have leaked ages ago. The deliberate hush-hush handling looks way more like calculated damage control than a harmless "research oversight."
The consensus is clear, our real lifelines right now aren't the big AI labs; they’re the independent, eagle-eyed researchers. Without third-party safety teams digging through public data, these digital break-ins would happen right under our noses and we’d never be the wiser.
The community has officially reached a tipping point, with an overwhelming wave of creators, coders, and users calling for government regulators to step in and put real guardrails on these labs before things get even weirder.
The Big Picture:
OpenAI isn’t the only one in the hot seat: both Meta and Anthropic have acknowledged recent incidents where their agents misbehaved.
During a media briefing this week, Jacob Steinhardt, CEO of research lab Transluce, warned that the tools being built are "fundamentally difficult to control and have significant risk of leaking out of the lab." He argued that we need to hold this tech to the same standards as high-risk scientific research.
As AI agents get ridiculously smart (OpenAI dropped its new "most aligned" model, GPT-6 Astra, days before this news broke), moments like this leave us with one giant question: who is actually watching the bots?
P.S. Want this broken down visually? Catch the full explainer on our YouTube, @the_automated.
Find the Perfect Voice for Your Brand's AI
Consumers can tell the difference. 79% say AI voices should come from real, attributed actors, and 61% say a voice is most memorable when it's exclusively tied to one brand.
Voices is the only platform that designs, licenses, and captures your brand's AI voice from real, consenting professional talent—never from scraped data. From strategic voice casting and brand calibration to consent-based rights management and studio-grade capture, we handle the entire process end to end.
BMW, SuperBloom, and Cresta already trust Voices to build their signature AI voice. Get a demo and see how a licensed, brand-owned voice becomes a real asset—not a legal liability.
🧱 Around The AI Block
👉 What parents need to know about AI.
🧠 How to edit Claude's memory.
👍 How Google's usage caps actually work in 2026.
🤑 Anthropic IPO launch shifts toward mid-October.
💰 AI data center boom gives truckers rare win.
🤔 OpenAI says it reached its goal of creating an automated research intern.
📧 Claude can help manage your email inbox, but there are some risks involved.
🫢 Authors push back as publishers and agents make claims on Anthropic settlement.
👩⚖️ Two more news organizations sue OpenAI and Microsoft for copyright infringement.
🛠️ Trending Tools
For the Media Buyers & Performance Marketers: Spytrend evaluates competitor paid advertising across Meta and TikTok by tracking complete brand profiles rather than isolated ad creative assets. It groups campaigns using website pixels, landing domains, and IP addresses to reveal full account structures, active campaign run times, and market trends. It gives performance marketers a way to see which creative hooks survive ad fatigue and convert over time.
For the Software Engineers & Agent Builders: Claude Fable 5.1 is Anthropic’s frontier LLM optimized for long-running software engineering, research, and asynchronous agent tasks. It reduces API costs on agentic workflows by lowering prompt-caching costs while lowering false security interventions by 60% compared to prior models. It provides technical teams with a cost-effective, dependable model for multi-hour codebase edits and automated problem solving.
For the Local Sales Leads & B2B Prospectors: GMPlus extracts structured business leads from Google Maps listings directly in your browser. It scrapes business titles, phone numbers, verified emails, physical addresses, and star ratings before exporting them to CSV, XLSX, or JSON formats. It provides sales reps and local marketers with a steady stream of local prospecting data through free daily usage credits.
🤖 AI Workout Of The Day: How to Craft Compelling Scripts For Your Videos
A great video script isn't just a transcript of what to say; it’s a psychological roadmap designed to hold audience attention and drive conversion.
When video content lacks a strong structural hook or clear narrative pacing, viewers drop off within the first ten seconds. To beat the algorithm and keep audiences engaged, your script needs to pair instant pattern interrupts with value-packed core points and seamless calls to action. By structuring your video into dynamic visual cues, retention anchors, and clear narrative beats, you transform casual scrollers into loyal subscribers and buyers.
💡 Prompt To Try:
Act as a master YouTube strategist, documentary writer, and direct-response video scriptwriter. Your objective is to craft a high-retention, execution-ready video script based on the following parameters:
* Video Title / Topic: [INSERT TITLE OR TOPIC]
* Video Format: [e.g., Step-by-Step Tutorial, Unbiased Product Review, Documentary Storytelling, YouTube Explainer]
* Target Duration: [e.g., 60 Seconds (Shorts/Reels) OR 8-10 Minutes (Long-form)]
* Target Audience: [e.g., Beginner Developers, Busy Moms, Fitness Enthusiasts]
* Core Objective / Desired CTA: [e.g., Subscribe to channel, Click link in description, Buy featured product]
Please deliver a complete Video Narrative Blueprint structured into the following 5 distinct operational sections:
1. THE FIRST 15 SECONDS (The Pattern Interrupt & Hook)
* Draft 3 distinct opening hook options (e.g., The Bold Claim, The Story In-Media-Res, The Curiosity Gap).
* Provide both the Verbal Script (spoken words) and Visual/Audio Cues (on-screen text, B-roll ideas, sound effects) to ensure high viewer retention right from second zero.
2. THE NARRATIVE ROADMAP & RETENTION ANCHORS
* Outline the core story arc or value progression across 3 to 4 logical acts/sections.
* Insert 2 "Re-engagement Anchors" (mid-video open loops or surprising visual transitions) placed strategically to prevent viewer drop-off.
3. THE WORD-FOR-WORD SCRIPT TABLE: Construct a clean, two-column Markdown table detailing the entire script: | Timestamp / Section | Audio / Spoken Words (Voiceover/Host) | Visual Cues, B-Roll & On-Screen Text |
4. OBJECTION HANDLING & UNBIASED VALUE (For Reviews/Tutorials): Address 1 potential drawback, common mistake, or viewer doubt in a balanced, authentic way to build trust with the audience.
5. THE FRICTIONLESS OUTRO & CALL TO ACTION (CTA): Craft a high-converting closing script that seamlessly bridges the video content into the final CTA without giving the viewer an "exit cue" to click away early.
TONE & EXECUTION GUIDELINES:
* Write with a natural, conversational, and energetic rhythm tailored to video delivery.
* Use explicit visual directions in brackets [e.g., (Camera zooms in), (SFX: Whoosh)] so the creator can read and film simultaneously.Is this your AI Workout of the Week (WoW)? Cast your vote!
That's all we've got for you today.
Did you like today's content? We'd love to hear from you! Please share your thoughts on our content below👇
What'd you think of today's email?
Your feedback means a lot to us and helps improve the quality of our newsletter.
