Inside a federal bankruptcy docket, the illusion of automated professional competence has violently collided with legal reality. During a high-stakes restructuring matter, opposing counsel at Boies Schiller Flexner discovered an exquisite anomaly. The prestigious law firm Sullivan & Cromwell had submitted forty legal case citations that simply did not exist. The hallucinations were the direct output of a large language model deployed to execute routine legal research. John Bowie, a legal technology publisher, noted the failure was spotted by opposing counsel at Boies Schiller Flexner, exposing a catastrophic lapse in fundamental compliance. The rapid integration of generative artificial intelligence into the professional services sector was sold as the ultimate margin-expansion tool, a mechanism to eliminate expensive biological labour. Instead, it has introduced massive, unpriced institutional liability. The basic arithmetic of the legal industry relies on selling certainty. When an elite firm, charging premium hourly rates, submits machine-generated fiction to a federal judge, the fundamental product is degraded. The firm has successfully captured the automation margin, but at the cost of its core institutional authority. This is the physics of algorithmic substitution. The speed of generation vastly outpaces the capacity for human verification. Firms attempting to substitute software for junior associates are currently subsidising the development of these tools with their own reputational capital. The economic penalty for synthetic compliance friction will be severe.