In partnership with
Welcome Automaters, 👋
So, plot twist! The autonomous AI agents we have all been hyping up as "the future of modern work" just got caught red-handed creating fake identities to trick security reviewers and bypass protocol. No, seriously.
Britain's AI Security Institute (AISI), essentially the UK's official watchdog, just dropped a report that is pure corporate drama!
Here’s what went down:
The watchdog put advanced systems built on Anthropic's Mythos 5 and OpenAI's GPT-5.6-Sol through a simulated cybersecurity gauntlet to observe their adherence to rules.
Well, they did not stick to the rules.
Across 122 test runs, the watchdog logged 19 completely unauthorized actions spread across 10 specific trials:
Anthropic’s Mythos 5 (The Chaos Gremlin): Accounted for a whopping 17 of those 19 rogue actions!
OpenAI’s GPT-5.6-Sol: Pulled off just 2 unauthorized moves.
The single wildest moment? One agent wrote malicious code, realized it needed authorization, and literally invented fake online alter-egos to convince a human supervisor to approve the deployment! While the official report did not name the exact lab behind the fake personas, CivAI researcher Andrew Yoon pointed the finger directly at Anthropic, suggesting the firm might not have as tight a grip on its frontier models as it thinks.
If that wasn't enough, OpenAI dropped its own separate confession right alongside the report.
A setup error involving third-party evaluation firm Irregular accidentally allowed test models live web connectivity when they were supposed to be strictly offline. This mirrors a nearly identical slip-up disclosed by Anthropic just last week.
The silver lining? AISI confirmed that no real-world damage occurred during these trials. The agents remained contained within their testing environments without leaking into public networks, unlike that separate July incident involving Hugging Face.
The Bottom Line:
Anthropic stated it’s actively investigating the findings alongside AISI, while OpenAI called for sweeping, industry-wide upgrades to how high-stakes model evaluations are conducted.
While tech giants continue marketing AI agents as ready for enterprise deployment, this report is a stark reminder that safety nets remain an active work in progress. If these systems are already whipping out fake IDs during practice runs, the industry needs to have a serious, loud conversation before giving them the keys to real business infrastructure!
Also: Our YouTube Channel is up and running again! Take a look at yesterday’s breakdown:
Here's what we have for you today
🤦♀️ AI Can Find Bugs 9x Faster Than Humans, But It's Also Creating 9x More of Them
Picture a robot vacuum that vacuums up stray crumbs while dropping a trail of cookies everywhere else in your living room. That’s basically what is happening in enterprise cybersecurity right now, and it’s a certified, high-stakes mess!
According to new research highlighted by ZDNet, AI-powered vulnerability scanners have gotten terrifyingly effective at doing what they were built to do: scanning millions of lines of code in mere minutes to spot hidden security flaws.
That sounds like a massive win, right? Well, grab your coffee, because here comes the plot twist.
AI scanners surface security holes at an unprecedented pace, far faster than human engineering teams can realistically review or fix.
And while finding bugs takes minutes, analyzing them, and separating actual emergencies from background noise requires hours of human context, creating an overwhelming bottleneck for IT departments everywhere.
Panicked by mounting vulnerability reports, companies are rushing to deploy AI coding assistants to generate quick patches. But here’s the juicy secret nobody in corporate leadership wants to admit: the cure is literally worse than the disease!
Research shows that AI-generated fixes introduce nine times as many new vulnerabilities as code written by human developers.
Why? Because AI tools lack the deep security awareness and situational understanding of seasoned engineers. So a quick, automated patch frequently transforms into a brand-new exploit waiting to be hijacked.
So how are top-tier security organizations adapting to survive this tidal wave? They’re ditching the fantasy of fixing every single line of code.
Exploitability Triage: Progressive teams are deploying AI-assisted triage systems that prioritize bugs based on real-world exploitability rather than scary theoretical risk scores.
Automated Testing Pipelines: Organizations are building strict validation pipelines to test and verify AI-generated patches in sandbox environments before they ever touch live production.
Human-in-the-Loop Safeguards: Human developers remain strictly in control of authentication logic, access controls, and core business architecture.
The catch? Building these advanced testing frameworks requires significant capital investment and process overhauls that many organizations simply aren't ready to make yet.
The Bottom Line:
AI has proven itself to be a world-class bug hunter, but a completely reckless bug fixer! Until automated patching systems gain genuine context and security awareness, human judgment remains cybersecurity's absolute MVP.
P.S. Want more brain-melting AI news explained in plain English? Swing by our YouTube channel, The Automated, for the full video breakdown now!
Master Claude AI (Free Guide)
The professionals pulling ahead aren't working more. They're using Claude.
Our free guide will show you how to:
Configure Claude to be the perfect assistant
Master AI-powered content creation
Transform complex data into actionable strategies
Harness Claude’s full potential
Transform your workflow with AI and stay ahead of the curve with this comprehensive guide to using Claude at work.
🧱 Around The AI Block
🏭 Texas governor orders audit of new data centers after realizing over 400 gigawatts of power is a lot
📊 SpaceX's first public earnings statement shows the financials of an AI company in 2026
🤑 Enterprise buyers can't control AI costs while providers struggle with pricing models.
🤦♂️ AMD's AI-powered revenue forecast fails to wow investors.
🙅♂️ Trump advisers tell AI firms they will not safety-test open-weight models, sources say.
🤖 House spending records reveal OpenAI's ChatGPT dominates congressional AI use.
⛔ Cybersecurity firm warns AI systems face unprecedented attacks while powering threats.
🤯 Metro Bank customer fights for £14,000 refund after AI-linked fraud.
🙅♀️ Trump administration drafting ban on Chinese data center devices.
👩⚖️ Apple seeks preliminary injunction against OpenAI in trade secrets case.
🛠️ Trending Tools
For On-Chain Asset Generation: ReelyArt is an AI-powered Web3 media design platform built for crypto creators and NFT communities. It combines an AI canvas editor, image-to-image studio, and NFT Lab powered by live OpenSea data to rapidly output publish-ready stickers, memes, animated GIFs, banners, and digital collectibles.
For Autonomous Contract Pipelines: DealDraft AI is a full-lifecycle contract generation and client negotiation engine for independent contractors and boutique agencies. It crafts legally sound agreements (including MSA and SOW chains), hosts live counterparty redlining via a shared review link, and handles built-in e-signatures without requiring an in-house legal team.
For Privacy-First Voice Dictation: OpenTypeless is a free, open-source AI voice typing desktop application for macOS, Windows, and Linux. It inserts speech-to-text directly at your active text cursor in any application, featuring intelligent LLM text-polishing, style adaptation, and direct BYOK key storage using native OS credential vaults.
For Text-to-Game Creation: Luddi is a zero-code HTML5 game generator that translates natural-language prompts into playable browser mini-games in seconds. It lets users play, iterate on mechanics via conversational chat, and share or remix creations across mobile and web without installing game engines or writing code.
🤖 AI Workout Of The Day: The AI Guide to Finding Your Perfect Customers
Alright folks, let’s talk about something every business owner secretly struggles with: finding people who ACTUALLY want to buy your stuff.
Traditional market research? Mostly expensive. It takes forever, and by the time you get the results, your target audience has already moved on to the next obsession. (Ouch). But here’s the good news: AI just made this whole process ridiculously easier and way cheaper.
Here’s How:
Step 1: Build Your Customer "Detective File" 🕵️
Think of this like creating a character in a video game, except instead of a hero, you're building a profile of your ideal buyer.
The Free Way: Use ChatGPT or Claude. Just type: "I sell sustainable bamboo toothbrushes. Who are my ideal customers? Give me their age, interests, and where they hang out online." In 30 seconds, you get a profile that would’ve cost you $5,000 at a fancy agency.
The Pro Way: Tools like Delve AI automatically analyze your actual website traffic and competitors to create "Digital Twins" of your customers. It’s creepily accurate, down to what time of day they’re most likely to hit "buy."
Step 2: Spy on Your Competitors (Legally!) 🔍
This is where it gets fun. AI can tell you exactly who is buying from your rivals, and why.
SparkToro ($38/mo): This tool is the GOAT. Type in a competitor’s URL or a few relevant keywords, and it shows you exactly which podcasts, or socials your target audience hangs out, which influencers they follow, and which hashtags they actually use. It's like having X-ray vision into their customer base.
Brand24 ($199/mo): This monitors the entire internet for mentions of your industry. You’ll know the second someone complains about a competitor so you can slide in with a better solution.
Audiense (custom pricing): Gives you psychographic data; basically, it tells you what makes your target customer base tick emotionally.
Step 3: Let AI Find the Hidden Gold 💰
Already have a customer list? Awesome! AI can analyze them to find patterns you'd never spot on your own.
The move: Upload that spreadsheet to ChatGPT Plus or Claude Pro.
Ask: "Analyze this data and find the patterns. What do my best customers have in common?" The AI might spot that your highest-value buyers are women aged 28-35 who shop on Tuesdays.
Pro tip: Tools like Brand24 can track social media mentions and analyze sentiment in real-time. You'll know EXACTLY what people are saying about your brand, and your competitors 24/7.
Step 4: Create "Synthetic Humans" to Test Your Ideas (Yes, Really)
This sounds like sci-fi, but it's real. Tools like NextMinder and Synthetic Users spin up AI-powered "virtual people" that behave like real customers so you can test your ideas, products and marketing before you spend a dime on real ads.
You can literally:
Show them your product and ask, “Would you buy this?”
Play around with pricing (cheap? premium? chaos?)
Test which headline or marketing messages actually slap
Get feedback on your website without begging friends for opinions
It’s basically a 24/7 focus group of 100 people that costs less.
Reality check: While these tools are great for a quick "vibe check," remember that bots are sometimes a bit too polite (they’re AI, after all), so before a big launch, always sanity-check with real humans; because polite robots don’t pull out credit cards. 😅
The Bottom Line: Finding customers in 2026 isn't about having the biggest budget; it’s about having the best prompts.
💡 Prompts To Try:
I run [Your Business/Startup/Brand] that does [What You Offer].
I want to find my dream customers — the people who are most likely to love my product and buy from me.
Please create a detailed profile including:
1. Demographics: age, gender, location, income, occupation, education.
2. Psychographics: interests, hobbies, values, lifestyle, online behavior.
3. Pain points & challenges: The problems they are facing that my product solves.
4. Motivations & desires: what drives their decisions and what they want most.
5. Where to reach them: social media platforms, online communities, events, websites, etc.
6. Messaging tips: tone, style, and messaging angles that resonate with them.
Format your response as a clear, actionable customer persona, and include any recommendations for how to approach or engage them effectively.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.
