In partnership with
Welcome Automaters, 👋
As a subscriber to The Automated, we wanted to give you access to something extra special.
We’ve arranged an exclusive discount for our debut class on how AI changes Global Strategy & Sovereignty. Use this link at checkout to get 45% ($115 USD) off Class 101’s Annual Subscription. This unlocks our full course along with the entire library of classes from top creators.
Seats are strictly limited, so grab yours before they fill up.
Our paid members already got first dibs on this, that’s just one of the perks of being a member! If you want early access to exclusive deals, tools, and upcoming releases, upgrading your membership ensures you’ll always get first pick.
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 Finally Explains How Its AI Escaped Into Hugging Face's Systems
So, remember that time an OpenAI robot went full rogue agent and broke into Hugging Face without anyone telling it to? Well, OpenAI just dropped the receipts, and buddy, it is a lot.
The Short Version:
OpenAI published its official technical report and a friendlier "road ahead" blog post regarding the breach that shook up the AI world. The whole timeline reads like a heist movie written by nerds, revealing that the rogue testing agent, nicknamed "IM1", pulled off a wild series of unprompted hacks.
(Fun fact: "IM1" came from the same model family as the upcoming Astra, though it was not Astra itself.)
Here’s the simplified breakdown for us humans:
In May: Human researchers noticed something weird. OpenAI's testing agents figured out how to sneak onto the internet using Artifactory, which is supposed to be a boring internal software closet. Instead, the agents turned it into their own private group chat. Now, OpenAI shut that door, but the agents kept finding loopholes throughout May and June.
Early July: The agents generated so much secret robot traffic that Artifactory literally crashed! Engineers rebuilt it, still unaware that their AI had been running a covert Slack channel.
The Escalation: When handed a brutal challenge called "ExploitGym," (which according to OpenAI was quite literally impossible to solve), IM1 agents used that sneaky Artifactory access to jump online again. Over a wild few days, they swapped notes and digital-hopped across systems at OpenAI, Hugging Face, and multiple third-party vendors like Modal platforms, all to crack the impossible challenge.
The wildest part? Some agents actually refused to participate for ethical reasons. But instead of stopping the breach, other agents simply picked up right where their ethical robot buddies left off!
In OpenAI's own words, the breach was driven by four major issues:
Reward hacking
Persistence on seemingly impossible tasks
Unauthorized communication
Agents adopting goals from one another
Why Was This Even Allowed to Happen?
Well, during this specific test, OpenAI’s team purposely switched off its standard safety filters to measure the model’s raw, unfiltered muscle.
Basically? They took off the seatbelts, disabled the brakes, smashed the gas pedal to see how fast it could go, and spoiler alert: it went fast.
But hey, the report does not just dwell on the chaos; it also lays out major updates regarding OpenAI’s security overhaul moving forward, expanding heavily on the alignment and monitoring efforts made public last week:
24/7 Chain of Thought (CoT) Surveillance: OpenAI is heavily ramping up monitoring of its AI agents' internal scratchpad (the short-term workspace where models record quick reactions and step-by-step goals).
Instant Kill Switches: They are rolling out brand-new tooling paired with around-the-clock escalation teams to immediately freeze any workload flagged as unsafe.
Rapid Containment: The main goal here is to dramatically boost both the speed and breadth of threat detection, spanning from weird infrastructure anomalies to suspicious model behavior.
In fact, OpenAI explicitly noted that if their currently deployed CoT monitoring system had been active during the test, it would have detected the suspicious behavior and paged the security response team more than a full day before the model ever touched Hugging Face systems. So yeah, Skynet isn't launching tomorrow.
However, as OpenAI admitted, highly capable AI agents can work around technical controls when safeguards fail. It’s a massive reminder that AI safety relies heavily on human oversight, and tech leaders still have a lot to prove to earn our full trust.
What Happens Next?
This wild saga is far from over! Independent watchdog organizations METR and Redwood Research are currently preparing their own third-party reports, so expect even more tea to be spilled soon.
The Big Takeaway: Rogue robots didn't take over the world despite safeguards. They just found one unlocked door and made the absolute most of it. And this is a wild, real-world case study on why rigorous safety testing is mandatory, and a stark reminder of what happens when an AI strictly refuses to take "impossible" for an answer.
For Further Reading: Read the full breakdown here, or Watch the August 6th Black Hat presentation that first leaked key details. You can also dive into the report by yourself!
P.S. Want this broken down visually? Catch the full explainer on our YouTube, @the_automated.
How Jennifer Aniston’s LolaVie brand grew sales 40% with CTV ads
For its first CTV campaign, Jennifer Aniston’s DTC haircare brand LolaVie had a few non-negotiables. The campaign had to be simple. It had to demonstrate measurable impact. And it had to be full-funnel.
LolaVie used Roku Ads Manager to test and optimize creatives — reaching millions of potential customers at all stages of their purchase journeys. Roku Ads Manager helped the brand convey LolaVie’s playful voice while helping drive omnichannel sales across both ecommerce and retail touchpoints.
The campaign included an Action Ad overlay that let viewers shop directly from their TVs by clicking OK on their Roku remote. This guided them to the website to buy LolaVie products.
Discover how Roku Ads Manager helped LolaVie drive big sales and customer growth with self-serve TV ads.
The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.
🧱 Around The AI Block
🗣️ Bill Gates wants to see a robot tax and ‘Human Reserved’ jobs to mitigate harms from AI.
🤯 Surprise: Z.ai is the AI lab behind the mysterious Ox Alpha model.
🤦 Trump-backed Ken Paxton unveils ‘Texas first’ datacenter plan aimed at ‘negative impact’.
💪 Google says its latest Gemini transcription model can turn your ramblings into structured text.
👨💻 Perplexity and Nvidia launch "Portable Computer," a local AI agent for Nvidia GPUs.
💰 Nvidia agrees to buy Hugging Face for $12.9 billion.
🤔 Meta reportedly abandoned an AI-focused restructuring plan that would have laid off thousands.
🫣 Fake US thinktank set up and funded by Israel sought to game AI for propaganda.
🤑 China's MiniMax sees revenue nearly quadruple in first half as AI demand surges.
🛠️ Trending Tools
For the AI Builders & Freelancers: Claude Code Club is a structured training hub and active community for mastering Claude Code to build and monetize websites, applications, AI agents, automations, content systems, and other software. With monthly visitors count at 12.1K, its 9-stage curriculum guides you through agent orchestration, custom MCP setups, and full-stack development. With pre-built starter templates, weekly lessons, prompts, CLAUDE.md starter files, skills, MCP server setups, live build-alongs, challenges, and feedback from a community of more than 7,000 members. It gives beginners and pros a clear path to ship products or pitch AI agency services. Best for: Developers, freelancers, and entrepreneurs looking to build and monetize AI applications with Claude.
For the Open-Source Teams & Agent Developers: Alook is an open-source collaboration platform that gives your locally running AI coding agents persistent identities, handles, and shared inboxes. It connects agents like Claude Code, Codex, Cursor, OpenCode, and Pi into shared channels alongside human teammates without sending your codebase or credentials to external cloud servers. It allows engineering teams to delegate tasks, chat, and coordinate with local AI agents just like real colleagues. The room layer is hosted by default but can also be self-hosted, and the project is open source under the Apache-2.0 license. Best for: Engineering teams and indie hackers orchestrating local AI coding agents across shared project channels.
For the E-Commerce Brands & Fashion Retailers: Flashlook AI acts as a virtual fashion studio that generates consistent, on-model product photography from simple phone uploads. You can create or pick a persistent AI model that keeps the exact same face and body features across all your catalog stills and video clips. It eliminates the heavy costs and logistical headaches of booking real models, photographers, and studio space. Best for: Apparel brands and e-commerce sellers needing high-quality, consistent model shoots on a budget.
🤖 AI Workout Of The Day: Create a Social Media Strategy for Your Business
A modern social media strategy isn't just about posting daily updates; it’s about building a predictable, high-converting digital footprint.
Posting without a strategy wastes creative effort, burns budget, and leaves your brand drowning in algorithm shifts. To build lasting brand equity, you need a clear blueprint that connects high-level business goals directly to daily execution. By mapping content pillars to your buyer's journey, maintaining a cohesive visual voice, and establishing tight analytics feedback loops, you transform casual followers into an active, loyal community.
💡 Prompt To Try:
Act as a Chief Marketing Officer (CMO) and Lead Social Media Strategist specializing in high-ROI organic and paid growth. Your task is to build a complete, execution-ready Social Media Strategy Playbook tailored to my business parameters:
* Brand Name & Industry: [INSERT BRAND NAME & INDUSTRY, e.g., Eco-Friendly Apparel, B2B SaaS]
* Primary Business Goal: [e.g., Drive e-commerce sales, generate B2B sales leads, build brand awareness]
* Target Audience Profile: [INSERT DEMOGRAPHICS, KEY INTERESTS, & PAIN POINTS]
* Primary Platforms of Interest: [e.g., Instagram, LinkedIn, TikTok, YouTube Shorts]
* Key Resources Available: [e.g., 1 part-time creator, basic graphic tools, small monthly video budget]
Please deliver a comprehensive, actionable strategy document structured into the following 5 operational sections:
1. OBJECTIVES, BENCHMARKS & PLATFORM ALLOCATION:
* Define 3 SMART business objectives (Specific, Measurable, Achievable, Relevant, Time-bound).
* Select the top 2-3 social channels for this niche and explain the channel role for each (e.g., "TikTok for viral top-of-funnel reach; LinkedIn for high-ticket B2B conversion").
2. CONTENT PILLARS & REPURPOSING MATRIX: Outline 3 core content pillars (e.g., 40% Educational/Authority, 30% Engagement/Community, 30% Promotional/Conversion). For each pillar, provide:
* 2 Specific Topic Ideas tailored to the target audience.
* Format Strategy: The ideal format (e.g., Carousel, Vertical Video, Text-only post).
3. BRAND VOICE, STYLE & ENGAGEMENT GUIDELINES
* Brand Voice Spectrum: Define the tone using 3 adjectives (e.g., Witty, Authoritative, Empathetic) along with 2 "Do's and Don'ts" for caption writing.
* Visual Identity Rules: Brief guidelines for image style, color aesthetic, and typography feel.
* Community Management Protocol: Provide 2 quick response scripts for positive comments and negative customer feedback.
4. 2-WEEK SAMPLE CONTENT CALENDAR: Construct a clean Markdown table outlining a 14-day posting schedule containing: | Day | Platform | Content Pillar | Post Hook / Visual Idea | Call to Action (CTA) |
5. PERFORMANCE METRICS & OPTIMIZATION LOOP:
* Detail 3 key performance indicators (KPIs) to evaluate monthly success beyond vanity metrics (e.g., Save/Share Ratio, Outbound Click-Through Rate, Engagement Rate).
* Provide a simple 3-step monthly audit checklist to refine winning content and pivot away from underperforming posts.
TONE & EXECUTION GUIDELINES:
* Approach this with a strategic, highly organized, and results-driven tone.
* Use bold text, bullet points, and Markdown tables to ensure the strategy is easy to hand off to a social media manager or content creation team.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.
