Cloud AI billing is usage-based charging for services such as large language models. Costs may depend on the number of requests, input and output tokens, chosen model, or supporting cloud infrastructure. Because usage can scale rapidly, a valid account or API key can incur substantial charges if compromised or misused.
In attacks involving hijacked AI access, an attacker may submit requests using an organization’s credentials, turning identity compromise into unexpected cloud costs. Defenders should restrict and rotate keys, apply least privilege, and monitor billing data alongside audit logs for unusual usage. Budget alerts can flag spikes, but they may not stop consumption; enforced spend caps, when available, can pause eligible usage at a configured limit. Billing records should also be checked to distinguish estimated, posted, disputed, or adjusted charges.