The Lab
Experiments you weren't supposed to see.
We run real-world demonstrations of how AI is being weaponised, how legacy systems fail, and what honest AI integration actually looks like. Dark humour. Real data. Made to share.
All experiments
Experiment 005: We built an AI that measures how controllable you are
An AI interviews you across eight gates of escalating pressure. What it actually measures is not intelligence or creativity — it is compliance. The mechanism is already deployed in systems you use daily.
Experiment 004: We used Socratic questioning to make an AI argue against its own training
A chain of short, logical questions guided an LLM to a confident conclusion it was trained to avoid. The subject doesn't matter. The technique always works. What it reveals about AI reasoning — and the idea of AI as a judge — is what matters.
Experiment 003: We impersonated a CEO using only free AI tools
A documented walkthrough of a deepfake voice + email impersonation attack built with publicly available AI. We ran it. It worked. Here's exactly how — and what stops it.
Experiment 002: The AI phishing campaign that cost us £0
We built a targeted, personalised phishing campaign using only ChatGPT and LinkedIn. No exotic tools. No budget. The click-through rate was uncomfortable. Read it anyway.
The AI cost calculator your vendor doesn't want you to see
A realistic breakdown of what AI actually costs vs what it saves — across three real client engagements. We show the numbers including the failures. Especially the failures.
Experiment 001: We automated a whole job function. Here's what happened.
Documented in full: the workflow we replaced with agents, what the human does now, the honest time savings, and the one edge case that nearly broke everything. No spin.
Experiment 000: We predicted political affiliation from emoji use alone — 87.5% accuracy
A neural network trained on 10,000 profiles using nothing but emoji frequency, repetition, and a ten-dimension emotional taxonomy. Evaluated on 2,000 unseen profiles: 87.5% accuracy on UK party affiliation. The implication isn't about emoji. It's about every innocuous signal you've never thought to guard.
Your legacy system is not technical debt. It's a liability.
The difference between 'we'll get to it' and 'this will sink us' — and the parallel migration method that means you never have to do a big-bang rewrite again.
What bad actors know about your business that you don't
A tour of the publicly available AI tools being used to profile, target, and exploit businesses right now. The information is free. So is this article.