Cost Harvesting
Cost harvesting is a type of digital attack where someone intentionally forces a company's AI to work much harder than it needs to, like someone repeatedly asking a store clerk complex questions just to waste their time and make the store pay for extra staff. The goal is to drive up the company's electricity or cloud computing bills until they lose money.
Cost harvesting is an adversarial technique targeting pay-per-use AI services by intentionally inflating operational expenses. Attackers achieve this by submitting high-volume, low-complexity queries or low-volume, high-complexity prompts designed to maximize compute, memory, or storage consumption. This pressure can trigger autoscaling, leading to significant financial impact and potential service degradation for legitimate users.
Cost harvesting is an adversarial exploitation of AI service billing models, characterized by the deliberate manipulation of request parameters to force the victim's infrastructure beyond nominal operating capacity. By leveraging high-volume, low-complexity queries, resource-intensive queries, or adversarial prompts that trigger excessive agentic resource consumption, the adversary induces artificial scaling events and saturates GPU/TPU pipelines. This results in direct financial loss through inflated cloud consumption costs and secondary denial-of-service effects via request throttling or latency degradation.