AAISM: Advanced in AI Security Management

By Isaca · Question Mode

QuestionQ123

AI Risk Management

An information security manager suspects that a model inversion attack is underway after detecting a high volume of queries with unusually specific, structured inputs sent to an image-generation AI. The outputs resemble images in the proprietary training dataset. Which option is the MOST effective long-term remediation?

Explanation

Limiting the detail available in model outputs reduces the information an attacker can extract through repeated, carefully structured queries, making reconstruction of sensitive training data substantially harder. Query-pattern blocking and source-based restrictions can be bypassed, while removing individual examples does not remediate the model’s general susceptibility to inversion.

Community Discussion

No comments yet. Be the first to start the discussion!