Incident Report Within roughly twenty days of letting its semiconductor engineers use ChatGPT, Samsung suffered three separate leaks of crown-jewel intellectual property, source code, defect-detection logic, and a recorded internal meeting, into an external system that learns from what it is given. The most technically capable company on earth then did something revealing: it did not try to recover the data. It could not. It banned the tool. This is the forensic anatomy of a governance failure whose defining feature is that, by the time anyone noticed, every remedy except prohibition was already too late.
human-in-control Why 'human-in-the-loop' has become the most comforting lie in AI governance - and how to fix it.
out-of-bounds Most teams red-team their LLM endpoints from the outside, firing adversarial prompts at a URL and watching what comes back. It works, until you ask what it actually proved. This is an argument for opening the box.
template A bill-of-lading-style disclosure template for handing off an AI model's residual risks downstream.
Provenance Artifact ยท Self-Assessment The book's central claim is that data lives in six dimensions, but most organisations govern only one (space) and confuse a single-dimension success for whole-problem control.
Observation It wasn't the model that failed - it was the chatbot. A banking assistant handed over every customer's credential with no jailbreak, no injection, no exploit. And that is exactly why your red-team tool will never catch it.
Stay Informed
Get notified when new incidents, essays, and artifacts are published.
No spam. Unsubscribe anytime. View our privacy policy.
