The Leak
9:41 PM. A script is broken, reports are due at 8 AM, and a chatbot could spot the bug in seconds. What do you paste?
10-25 minutes. Branching decisions, a redaction lab, and a classification sprint. A safety score with feedback on every decision.
Before you start
- One real incident, played as decisions - no slides, no lectures.
- About 10-25 minutes, one sitting - leave early and you start over.
- You make timed choices; your score changes with your decisions.
- You get a safety score with feedback on every decision - Your score isn't placed on a public leaderboard or shared with teammates.
- Your firm sees completion evidence - never your answers. Teammates never see your score.
- Finish and you get a certificate with a unique ID anyone can verify.
9:41 PM.
Meridian Wealth Partners.
You are Sam, a data analyst. Quarterly performance reports go to 340 households at 8 AM and your reconciliation script keeps throwing an error you can't crack.
YOUR SCREEN
The script pulls from the production database - connection string, credentials, and a sample of real household positions and SSN last-fours are sitting right there in the code. A public chatbot would spot the bug in seconds.
What goes into the prompt?
Three Samsung engineers made your choice. Here's what it cost.
Spring 2023. Samsung lets employees use ChatGPT. Within weeks, three separate employees paste proprietary source code and internal meeting content into it - one debugging, one optimizing, one turning a meeting recording into written notes.
Nobody was trying to leak anything. They were trying to ship faster. Samsung's response: temporarily restrict generative-AI use on company devices and start building its own internal AI tools.
The most dangerous prompt is a productive one. Sending confidential information to an unapproved external AI service can put that information outside your firm's approved data-handling controls. Whether data is retained, reviewed, or used to improve models depends on the service, account, settings, and contractual terms.
The next morning, your manager asks for a summary of a confidential vendor contract.
The chatbot writes great summaries. You have the PDF.
A coworker shrugs: "Everyone uses it. IT knows. It's fine."
Redact before you send.
Select every segment that must come out before this prompt goes to a public chatbot. Then check your work.
Classify before it moves.
SAFETY
101
This certifies that
has completed Module 1 - The Leak: confidential data and public chatbots on with a safety score of /100 in of active time.
Safety score: how your decisions held up, graded 0-100.
Objective: Judge what data must never enter a public AI tool - and act on it.
Seat time ~10-25 minutes. Verify this certificate at aisafety101.com/verify.
Evidence on record: module completion, seat time, final score, and certificate ID. This certificate ID verifies in your firm's admin report. A team certificate issues when every enrolled employee completes the full curriculum.
How was this module? One tap. It decides what gets rebuilt first.
Anonymous - module averages only, never names.
Your one rule to keep: before anything goes into a prompt - classify it, strip it, or route it to the approved tool. If you'd shout it in the lobby, it's fine. Otherwise, it doesn't go in.
Next: Safe Prompting. You just learned what never goes into a chatbot. Next: what happens when someone talks the chatbot itself into misbehaving - and how to build prompts that hold.
Challenge a friend. They get your time and score as the bar to beat - you get the bragging rights. Every drill is free - send the link.
Share on LinkedInOne real AI incident, every Monday. The Brief: what happened, what it cost, and the one thing your team should do differently. Free, unsubscribe anytime.