CyberCertPrepAI Security FundamentalsSecure AI Deployment & MLOpsEASYFree question
Why must a model fine-tuned on one customer's data not be served to a different customer in a multi-tenant AI platform?
- A.Fine-tuned models are always too slow for other customers
- B.Licensing terms forbid running any model more than once
- C.The fine-tuned weights can memorize and regurgitate the first customer's confidential training data to the second customer
- D.Different customers always need different programming languages
Why C is correct
Fine-tuning bakes customer data into the weights, and language models are known to memorize and emit training examples, so serving tenant A's fine-tune to tenant B is a direct cross-tenant data disclosure path. This is why multi-tenant platforms bind each fine-tuned artifact to its owning tenant. Speed, single-use licensing, and programming languages are not the issue; weight-level memorization is.
Know someone studying for AI Security Fundamentals? Send them this one.