Algorithmic audits can reveal harmful biases, but what happens when an auditee tries to hide its decision-making?