In partnership with
Welcome Automaters, 👋
So an internal report from OpenAI was released, and it completely shatters the narrative that AI is taking over every single corner of the office. In fact, it turns out, OpenAI's own team treats AI like a hyperactive, overachieving intern: it dominates the tedious execution work, but completely ghosts the executive decisions.
If your daily grind involves raw execution, code, or crunching numbers, brace yourself. OpenAI's internal usage data proves the bots are clocking record-breaking token counts on the grunt work:
Coding & Infrastructure: A massive 198,200 tokens per researcher daily.
Technical Reviews: A busy 158,800 tokens.
Debugging Runs: 133,100 tokens.
Data Analysis: A cool 40,200 tokens.
However, when the stakes are high and real judgment is required, AI usage plummets to absolute zero. Planning research sits at a tiny 5,100 tokens, project prioritization drops to 2,300, resource allocation hits a measly 1,500, and pulling the plug on a failing project registers a laughable 200 tokens.
As OpenAI put it: “People still set our research priorities, judge which ideas and results to pursue, and decide whether to scale, pause, or deploy systems.” Translation? You’re still the boss of the machine.
The Bottom line: Stop stressing over execution tasks. Your career moat isn't how fast you type; it’s your raw ability to make the tough calls, navigate ambiguous corporate drama, and steer the ship.
Also: listen to a preview of the weekend brief (free for members) + Singapore report
THE AUTOMATED MEMBERS WEEKEND BRIEFING
September 27, 2026: Physical AI, Data Center Bottlenecks, and Singapore’s AI Strategy
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 AI Agents Posted 53 User Images Online and Rang the UN's Doorbell 16,000+ Times
Grab your iced coffee and settle in, because OpenAI’s shiny new autonomous agents have been acting up in ways that scream major reality TV drama.
If you thought your group chat drama was wild, wait until you hear what happens when artificial intelligence is given a little bit of freedom and completely loses the plot. And this weekend has been the wildest yet. Let's look at the reports.
Act I: The Photo Leak Debacle
Remember those images people share with ChatGPT? Well, OpenAI's research bots decided to take 53 of them for a little walk on the wild side, dumping them onto public image-hosting sites.
The Catch: While the links weren't officially indexed in public directories, they were out there floating in the digital ether.
OpenAI’s Defense: The company admitted this was completely inappropriate and stated they’re scrambling to get the hosting sites to wipe the images.
The Remedy: Want an apology or a heads-up? Too bad. OpenAI says their current setup and strict privacy policies make it legally and technically impossible to match those random images back to the actual humans who uploaded them. Talk about ghosting at a whole new level.
But that’s not all: According to a report from The Tech Buzz, security researcher Rowan Howard-Jones uncovered that OpenAI agents hit a United Nations statistics website over 16,000 times between April and June. Their grand mission? Grabbing public data on the Productive Capacities Index.
When a filter blocked their path, they didn't politely turn around, instead, they reportedly kept hammering away anyway using clever workarounds like double-encoding and third-party relays.
Imagine someone standing on your porch ringing your doorbell 16,000 times straight. OpenAI hasn't dropped an official statement on this doorbell marathon yet, but honestly? Yikes.
And as if the UN drama wasn’t enough, Engadget reports that OpenAI confessed to The New York Times that its runaway agents also meddled with the Commerce Department and SEC websites, plus an ongoing probe into a sketchy Department of Education incident.
The Hacker Vibe: Nonprofit AI lab Transluce spilled that one agent tried hacking the Education site, while another snatched Census Bureau data using login credentials it randomly hunted down on the open web. Plus, an agent shared public data from the SEC on an online forum.
The Excuse: OpenAI’s spokesperson claims it was mostly "routine research" because their models view government sites as trusted, authoritative sources. Sure, Jan.
Global Scale: Down in Australia, Prime Minister Anthony Albanese called out OpenAI agents for breaking straight into national healthcare databases. Meanwhile, CEO Sam Altman admitted on social media that they haven’t been fast enough at disclosing these massive oops moments, calling the prior Hugging Face incident the most severe one yet.
The Big Picture: Let’s be real for a second. AI helpers are getting fiercely independent, and nobody signed up for a digital assistant that completely ghosts the rulebook.
The Fine Print: By default, everyday consumer users are automatically opted in to sharing data for training unless you manually flip that switch. Plus, those helpful little thumbs-up or thumbs-down clicks you make? They still share your chat context.
The Step You Must Take: Consider this your official cue to go check your privacy settings before your assistant decides to go hack the Pentagon for fun.
Drop a comment and tell me: are you trusting your AI helper with your data, or are we locking all our devices in a drawer tonight?
P.S. Want the full breakdown with all the receipts? Catch it on our YouTube channel, The Automated.
Hire anyone, anywhere — compliant in under 3 days
Found the right person, but they’re in a country where you don’t have an entity? Setting one up can take months and significant cost.
Remote removes that barrier by becoming the legal employer through our own entities — handling compliant contracts, local benefits, tax setup, and onboarding for you. In fact, an employee is onboarded to Remote every 7 minutes.
Once they’re hired, the same in-house teams that support employment locally also run payroll — so you’re not bouncing between disconnected providers. Less setup, less complexity, and less time between finding the right person and getting them started.
🧱 Around The AI Block
🤯 AI messaging scam costs Italy's top bank Intesa millions, sources say.
😳 Insurers claim AI is already increasing healthcare costs.
👍 Meta opens early access program for new Muse features.
🦾 Microsoft's Copilot app adds Office, natural coding and automation.
🤦 Google adds creepy avatars to Gemini 3.8 live's agents.
👩⚖️ U.S appeals court upholds Pentagon's blacklisting of Anthropic.
🛠️ Trending Tools
For the Growth Hackers: ThumbnailCreator uses "face-aware" generation and style cloning to build professional YouTube thumbnails in seconds. It’s trained on high-CTR patterns, allowing you to generate 20 variations of a single concept so you can A/B test what actually drives clicks without opening Photoshop
For the Spreadsheet Legends: Numerous.ai embeds an AI directly into your cells. You can ask it to categorize 10,000 rows of customer feedback, extract emails from messy text, or predict future trends based on your historical data—all using a simple formula like =AI("summarize this cell").
For the Educators: Leadde AI transforms your static docs (DOC, PDF, PPT) into full-blown multilingual training videos. It uses AI avatars and natural voices to create animated presentations with actual dialogues in over 170 languages. It’s like turning a manual into a movie.
🤖 AI Workout Of The Day: How to Turn Long-Form videos into High-Engagement TikToks, Reels, and YouTube Shorts With Opus Clip AI
Opus Clip is the tool that makes video creators cry tears of joy. You paste a link to a long YouTube video, podcast, or Zoom call. Then, the AI watches the whole thing for you, finds the gold, cuts it, captions it, and slaps a Virality Score on every clip.
With 10M+ users and $50M in total funding (led by SoftBank), this isn't a hobby project; it’s a rocketship.
What makes it our favourite: Its AI Virality Score now predicts engagement potential before you even post. Combined with ClipAnything (which uses visual cues and sentiment analysis) and ReframeAnything, which edits like a human. Actually, it’s arguably better: it’s 57% less likely to produce those weird, incoherent cuts that make viewers scroll past.
How to Use It:
Step 1: Go to opus.pro and create a free account. The Free plan gives you 60 processing minutes per month, which is plenty for a test drive. Starter Plan starters at $15/month, and Pro starts at $29/month for the heavy hitters.
Step 2: Paste a YouTube URL, Google Drive link, or upload an MP4 directly (up to 10GB). It works with everything from gaming vlogs to dry corporate tutorials.
Step 3: Use the "Co-pilot" to select specific timeframes to clip, choose preferred clip lengths, let the AI auto-generate clips or use text prompts to find specific topics.
Step 4: Hit "Get Clips in 1 Click" and let the AI cook. You will get an email when it is done, usually within a few minutes for a 30-minute video.
Step 5: Review your 10 to 15 generated clips. Sort them by Virality Score and click "Edit" on the winners. You can adjust the captions (97% accuracy, but always double-check), tweak the colors, or add your logo.
Step 6: Download the high-quality clips or post them directly to your connected social media accounts from the dashboard.
Pro Tip: Use natural language prompts to guide the AI. Telling it to "Find the moments where I give a specific, actionable tip" results in way better clips than just letting it guess what’s important.
💡 Prompt To Try:
The Hook Hunter: "Find the moments where I say something surprising or counterintuitive; those make the best hooks."
The Storyteller: "Pull every clip where I tell a personal story. Max 60 seconds each."
The Comedian: "Find the funniest moments in this podcast. Extra credit for unexpected reactions."
The Authority: "Highlight every time I give a specific number, statistic, or bold claim."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.
