Stratagems #19: Mark Found His AI Audit Method in a Training Manual. He Left a Trap in His Report.

Short summary

Mark, an AI auditor, discovers someone has reverse-engineered his audit methodology and is using it against client SynthData's AI pipeline. A former employee's orphaned IAM account runs a weekly cron job that offsets a model-accuracy probe's time features by one hour, causing false-positive deviation alerts that trigger automatic model rollbacks every three days. The sabotage is invisible because the cron reverts itself within two minutes, and the model recovers when timing aligns again.

• •Orphaned IAM account 'ops-deploy-02' edits and reverts a cron job weekly in under two minutes
• •Cron offset shifts time-dependent features in a model accuracy probe, triggering false deviation alerts and automatic rollbacks
• •Saboteur used Mark's own AI audit playbook against him, indicating methodology leakage

Generated with AI, which can make mistakes.