Backdoor Attack

An attack that embeds a hidden trigger in a machine learning model during training, causing it to produce attacker-specified outputs when the trigger is present in the input.

Hiding a secret trigger in an AI during training. The AI works normally 99% of the time, but when it sees a specific hidden signal, it does something the attacker wants; like always approving their transactions.

Backdoor attacks are a supply chain risk that your AI governance framework must address. Organisations using pre-trained models or third-party AI components should implement verification and testing procedures to detect hidden backdoors.