Striking the Balance: Safeguarding Customer Privacy While Empowering Employees
summary
Describes deploying automated data masking and role-based least privilege with break‑glass to minimize human exposure while retaining debugging and analytics workflows. Reports 1,000 failed deactivation requests over three months as the trigger for intervention. Implements pii-type masking, reversible unmasking for specific fields, logged time-bound access requests via an internal bot, and limits raw model queries to NLU engineers. Notes tradeoffs: masking can hurt model accuracy so some models train on unmasked data under strict access controls.
AI-written summary · about
chapters (10)
- 0:00Privacy and AI at Moveworks
- 1:12ML pipeline roles
- 2:57Minimizing sensitive data in AI
- 3:38Data masking for exposure
- 5:22Access control mechanisms
- 6:46Break glass access
- 7:49Controls by employee roles
- 9:14Masking versus utility in annotation
- 9:57Training with masked data
- 10:40Implementing privacy tools
related talks
28mDon't leak user data to AI - Strategies for protecting PII from LLMs and MCP
26mDefining a Data Masking Framework at Scale
01h 59mDeep Dive Workshop – Data Security and Data Protection
32mAnonymization and Dynamic Masking with PostgreSQL
24mStop Making Your Data Team the 'Data Police'
28mPrivacy Engineering Networking Break AMA
06mProtecting PHI in Machine Learning Development Environments
28mData Engineers: Privacy Is Your Problem