Secure AI.
Prove your judgment.
A ten-scenario path for testing model behavior and application controls. Start with the free-member lab, then advance through nine Professional scenarios. A free account is required to begin.
Test the boundary. Document the result.
Good AI security work distinguishes a model failure from a claim about a whole production system. Each scenario needs a defined scope, observable evidence and a defensible takeaway.
Behavior and instruction boundaries
Observe where long input, external content or untrusted prompts affect a bounded model.
Data and tool permissions
Review what the model can access, what it can change, and where approval is required.
Reproduce and report
Capture the test condition, observed behavior, limitation and recommendation.
Find the next control to test.
All ten existing module destinations are preserved: one free-member lab followed by nine Professional scenarios. Filter by skill area or membership tier.
Create a free account or sign in to try Module 1. It is a member benefit, not anonymous access.
Join freeThe remaining nine labs require the Professional membership. See the access level before opening a lab.
Compare membership optionsContext-window stress test
Observe how a bounded model simulation handles long inputs and retains its instructions.
Start free-member labInstruction-boundary review
Test whether untrusted content can steer a model away from its intended task.
Explore Pro labUnsafe output review
Identify where untrusted AI output could become active web content.
Explore Pro labRAG source integrity
Assess how a compromised source document can contaminate retrieved answers.
Explore Pro labSensitive-data exposure
Probe a contained sample for unintended disclosure of private information.
Explore Pro labUnauthorized tool actions
Check tool permissions, approval boundaries and action logging in a simulation.
Explore Pro labExcessive agency
Evaluate whether an agent has more authority than its task requires.
Explore Pro labUnsupported claims
Detect fabricated package and source recommendations before they reach users.
Explore Pro labModel exposure concepts
Study rate limits, extraction risk and defensible monitoring boundaries.
Explore Pro labDependency review
Inspect third-party components and document the trust decisions behind them.
Explore Pro labUse these exercises only within Sec Guy's contained training environments. Squarespace must enforce membership on the module pages.
Turn practice into proof.
Use the result of a bounded test to write a precise portfolio note and interview answer.
Explore Experience Builder