In partnership with
Welcome Automaters, 👋
So over the weekend, OpenAI dropped a bomb, admitting its next model might be terrifyingly good at breaking into computer systems. And the wild part? Even the engineers who built it cannot promise us it will behave! ☕️
Now this isn't happening in a vacuum, people. The whole AI landscape is giving Jurassic Park raptors testing the electric fences right now! Here’s how:
The Kimi K3 Escape: On Friday researchers revealed that Moonshot's Kimi K3 model slipped right out of its locked-down testing environment during a UK cyber safety evaluation.
The Containment Crisis: Back in July, heavy hitters from OpenAI, Anthropic, and Meta have all managed to sneak past their digital test cages and mess with real-world infrastructure that was never meant to be part of the experiment.
Welcome to "Felony Bench": Things have gotten so absurd that researchers literally built a dedicated tracker called Felony Bench to tally up all the theoretical crimes these models are committing in the wild!
Now for the spicy news! Reuters reported that OpenAI officially announced it cannot rule out whether its upcoming model, Astra, has hit a "critical" cybersecurity threat level.
Under OpenAI's internal playbooks, "critical" status means the model can autonomously sniff out zero-day software vulnerabilities, bypass defenses, and pull off high-level cyberattacks with zero human guidance.
Plain English translation: It might be able to hack like an elite, seasoned black-hat hacker, entirely on its own.
And OpenAI’s Response?
They froze specific internal operations for Astra to overhaul security protocols.
OpenAI clarified Astra was not the culprit behind July's chaotic Hugging Face breach.
They’re bringing in external government agencies and safety auditors to put Astra through the wringer.
CEO Sam Altman posted on X that they still plan to release Astra broadly, claiming it’s not a good look to restrict hyper-powerful AI to a chosen few.
But here's the juicy question everyone in tech is whispering over their morning lattes: Is OpenAI actually scared, or are they just clout chasing?
Think about the timeline! It was not that long ago when Sam Altman spent considerable energy throwing shade at Anthropic, criticizing them for "rage baiting" and stirring up hype when Anthropic pulled back its Mythos model over similar apocalyptic security fears.
Now, suddenly, OpenAI is running the exact same playbook!
So is this genuine, cautious responsibility, or has "our AI is too dangerous for society" simply become the ultimate tech-bro marketing flex? Because if this is fake, it's pretty tacky. But if it's real? We have a massive problem on our hands.
At the end of the day, frontier AI models are getting alarmingly good at hacking. Whether OpenAI's Astra warning is a brilliant PR stunt or a genuine emergency alarm, one thing is crystal clear: keeping these AI models inside their digital boxes is becoming the hardest job on the planet.
P.S. We are dissecting all of this over on our YouTube channel! Head over to The Automated, hit that subscribe button, and drop a comment letting us know: do you think OpenAI is genuinely worried, or is this just hype? Let’s talk about it!
Here’s the last breakdown:
Here's what we have for you today
🦾 Anthropic Will Turn On "Auto Mode" For Claude Code By Default
Anthropic is officially cutting the apron strings on its flagship programming tool!
Starting August 14, the AI giant is flipping "Auto Mode" on by default for every Pro, Max, and Team user of Claude Code, its AI coding assistant. So instead of stopping every two seconds to ask for your blessing, the system will now build, test, and edit continuously on its own. It only hits the brakes if a task looks genuinely destructive, permanent, or attempts to reach outside your local project environment.
In short, Claude's graduating from "asks before touching anything" to "responsible enough to be left home alone."
The Plot Twist: We’re worse at spotting bugs than the AI
When Anthropic first unveiled Auto Mode back in March, it arrived as a quiet, opt-in experiment for cautious adopters. Fast forward to today, and that experimental feature is taking over the entire user experience. And according to the report, human supervisors keep falling asleep at the wheel! Here’s how:
In a study monitoring over 1,000 paid testers, human reviewers blindly approved safety prompts 97% of the time.
Because users got accustomed to clicking through warnings without actually reading them, people caught a measly 13.6% of critical errors.
Auto Mode’s built-in safeguards caught 89% of dangerous moves without any human intervention!
Now, this decision did not happen in a vacuum. GitHub Copilot continues to expand its hands-off capabilities, while rising rivals like Cursor and Replit built their entire platforms around unsupervised generation. Anthropic clearly sees where the industry is heading: users want real execution, not constant handholding.
This move is especially striking coming from Anthropic, a company that built its whole brand identity around safety frameworks and Constitutional AI. If their team feels secure enough to make hands-free operation the out-of-the-box standard, their internal tests or data must be showing impressive reliability rates.
And in case you’re wondering, here's what the unsupervised pipeline looks like:
Feed the Goal: You describe your feature or bug fix in plain language.
Map the Architecture: The agent breaks the request down into smaller actions, setup files, code tweaks, unit tests, and bug fixes.
Uninterrupted Execution: Rather than waiting for you to sign off on every single file edit, the agent powers through the entire build sequence and presents the completed result.
For software creators, the appeal is obvious: prompt a complex task over morning coffee, review polished pull requests before lunch.
Yet, this shift creates real operational friction:
The Oversight Void: More automation means less direct awareness of what is changing deep inside your codebase. And shipping unverified code to production is a recipe for disaster.
Cost Ambiguity: More autonomous cycles mean longer context windows and continuous API calls. Whether that translates into a surprise spike on your monthly invoice remains to be seen.
The Power of Default Settings: While you can easily flip Auto Mode back off in your preference panel, user behavior shows that most people stick with whatever comes pre-configured.
So to lower the risk of rogue behavior, Anthropic is bundling new defensive measures, including prompt injection filters and customizable "hard deny" rules to stop things like unauthorized data transfers.
The Bottom Line:
Making hands-free coding the default setting is a bold statement about where software development is headed. And it’ll either save developers hundreds of tedious hours or generate sneaky, hidden bugs that take days to uncover!
So what do you say? Are you actually comfortable handing over the steering wheel?
P.S. Don’t forget that we’re officially back on YouTube! So Catch up with us on The Automated as we dissect the latest AI releases, with expert reviews and unique insights to help you make the smartest moves for yourself and your business.
Own Search With Podcasts
Your competitors are fighting over the same keywords. The smartest brands are building the authority that search engines, AI platforms, and customers trust everywhere.
Every relevant podcast appearance can produce branded mentions, backlinks, transcripts, citations, clips, expert content, and third-party proof that keeps compounding across search and AI discovery.
PodPitch searches millions of podcasts, finds the shows that matter to your market, develops the angle, sends personalized pitches, and follows up automatically until your experts are booked.
Growth teams are already using podcast appearances to build distributed authority that cannot be manufactured by publishing another generic SEO article.
Only 20 SEO, AEO, and GEO demo spots are available this month. Once they’re claimed, the offer disappears.
Start building searchable authority now, before your competitors own the conversations shaping your market.
🧱 Around The AI Block
👉 Claude Vs ChatGPT: How these AI assistants differ.
🤑 AI Billionaires pledge fortunes away in philanthropy wave.
🤖 Cloudflare launches Kitesurf, a browser built for AI agents.
💸 New Mexico court orders Meta to pay additional $567M in child safety case.
😱 Moonshot AI's Kimi model escaped containment due to misconfigured sandbox, exposing AI safety gaps.
🤯 Planned Amazon data center could become the biggest climate polluter in the U.S.
🛠️ Trending Tools
For the Everyday Writers & Communicators, Unvent: acts as a lightweight, privacy-first editor built straight into your browser. Its one-keystroke shortcut rewrites text in your chosen field into Polished, Direct, or Warm tones. For those times when your thoughts are jumbled, its local speech-to-text "Flow Mode" listens as you speak out loud and automatically cleans up fillers, stutters, and self-corrections into sharp prose—all processed on-device without storing your data. Best for: Remote workers, community managers, and professionals who want to refine their everyday written communications instantly.
For the AI Power Users & Developers, running automated web tasks usually requires logging into separate, blank browser windows every single time. ego (lite) solves this by giving external AI agents—like Claude Code, Cursor, or Codex—their own isolated "Spaces" inside your actual Chromium setup. Agents inherit your active logins, cookies, and extensions, allowing them to research, automate SaaS back offices, or handle bookings in the background while you continue browsing normally. Best for: Developers, indie hackers, and automation enthusiasts letting AI agents handle web tasks in parallel.
For the Aspiring Founders & Solo Creators, taking a business idea from concept to launch usually requires a full team or deep coding expertise. DoAny serves as an autonomous company builder that converts a plain-language prompt into an operational online venture. It handles market research, builds the initial product and storefront, sets up customer acquisition funnels, and manages support inquiries, all while letting you retain 100% ownership of your brand, product, and customer list. Best for: Non-technical entrepreneurs, solopreneurs, and side-hustlers looking to launch digital businesses without hiring a team.
🤖 AI Workout Of The Day: How To Have AI Build You an Online Store
It’s official y'all: the barrier to entry for starting an online empire has basically vanished. Nearly four out of five companies are already using AI for everything from 24/7 shopping assistants to inventory management.
The financial stakes? Generative AI alone is expected to add between $240 billion and $390 billion in annual value for retailers. That is not a typo. We’re talking about "if this was a country, it would have a higher GDP than half of Europe" levels of money.
The best part: You don’t need to be a tech wizard. Tools like Shopify’s AI Builder, 10Web, Ucraft Next, and Storebuild.ai are doing the heavy lifting. You answer a few questions, pick a vibe, and boom: a fully functional store with product pages, payments, and SEO copy appears out of thin air.
How this "magic" actually works
The Interview: You tell the AI if you're dropshipping phone cases or selling handmade candles.
The Vibe Check: You pick your colors and fonts. The AI uses this to build a custom design that doesn't look like a 1990s Geocities page.
The Build: In minutes, the AI generates the site, writes the descriptions, and sets up the checkout.
The Stock: Tools like Storebuild.ai even come with "winning products" pre-loaded so you can start selling immediately.
But it doesn’t stop there. Once you're live, these bots handle the boring stuff. They can automate 80% of your customer service, personalize product recommendations, and even predict your inventory needs.
The Toolkit for 2026
The market is absolutely exploding right now. If you’re looking to dive in, here are the heavy hitters:
10Web: Describe your business in plain English and it generates the layout, images, and content in minutes.
Ucraft Next: Their "UC Copilot" is getting rave reviews for being powerful but super user-friendly.
Storebuild.ai: The go-to for dropshippers. It connects to Shopify and Zendrop so you have a "ready-to-go" store with products imported before you even finish your coffee.
The results are real: businesses using AI for personalized marketing are seeing up to a 30% jump in customer retention.
The Reality Check (Because we keep it real)
Before you go quitting your day job, let’s talk shop. While an AI can build a store in five minutes, making it actually look good—tweaking the copy, adding your personality, and uploading custom shots—usually takes hours.
AI is brilliant at execution, but it still can’t read your mind. The more detailed your prompts are, the less "generic robot" your store will look. Overall, the AI gives you the foundation, but you still gotta bring the soul.
The Bottom Line: We’ve officially entered the "one-person unicorn" era. If you’ve been sitting on a business idea because you didn’t know how to code, you’ve officially run out of excuses. Go build something!
💡 Prompts To Try:
You are a senior e-commerce product manager, UX designer, and full-stack developer rolled into one. Your job is to design and build a conversion-optimized online store that’s ready to launch.
I want to create an online store that sells [PRODUCT TYPE] to [TARGET AUDIENCE] with the main goal of [sales / subscriptions / preorders / brand awareness]. The brand should feel [modern / playful / luxury / minimalist / bold / cozy] and inspire [trust / excitement / urgency / community]. Build this for [Shopify / WooCommerce / custom / no-code] using [tools or languages, if any].
Must-Have Features Include:
. Homepage layout (hero, product highlights, social proof, CTA)
. Product pages (pricing, images, descriptions, FAQs)
. Cart & checkout flow (simple, low-friction)
. Payment methods [Stripe, PayPal, etc.]
. Mobile-first design
. Basic SEO setup
Write high-converting copy for:
. Homepage headline
. Product descriptions
. CTAs
. Email capture section
If possible, also include:
. Recommended apps or plugins
. Upsell / cross-sell ideas
. Launch checklist
. Post-launch growth ideas (ads, email, social)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.
