Model Stealing
An attack where an adversary attempts to replicate a proprietary machine learning model by querying it and using the responses to train a substitute model.
In plain language
Copying someone else's AI by asking it questions repeatedly and using the answers to build your own version. Like reverse-engineering a secret recipe by ordering the same dish many times.
Why this matters
Model theft is a competitive and intellectual property risk. Your governance framework must include controls to protect proprietary models through access restrictions, query monitoring and legal safeguards.
Relevance
GovernanceIP protection controls for proprietary AI models are a governance requirement that must be specified in your security and risk management policies.
Related terms
Putting model stealing into practice in your organisation?
Ready to transform your AI strategy?
Partner with Australia's AI strategy and governance specialists. From adoption roadmaps to ISO 42001 audit readiness.