Organizational siloing is a coordination problem. It is also, simultaneously and more precisely, a legal architecture — one that determines whose knowledge counts as institutional knowledge and whose knowledge is safely contained.
Every large regulated organization operates through functional siloing: quality assurance is separate from production, legal is separate from operations, research is separate from product development, trust and safety is separate from revenue. This structure is universally described in organizational literature as a coordination mechanism — specialization enables expertise, clear functional boundaries enable efficient resource allocation, and separation of duties reduces error. These benefits are real and are the dominant explanation for why siloing persists across industries and organizational types.
This paper argues that organizational siloing in regulated industries performs a second function simultaneously, and that understanding this second function explains features of silo design that the coordination explanation cannot account for. The second function is legal: the silo determines which employees' knowledge constitutes institutional knowledge for purposes of regulatory and legal liability, and which employees' knowledge is contained within units whose discoveries do not bind the corporation. This is not a bug of the silo structure — it is a feature that is designed, maintained, and in some cases explicitly managed as a legal strategy.
The clearest evidence for the second function is not in the internal communications of corporations that have siloed strategically — those communications are typically privileged or destroyed. The clearest evidence is in the pattern of what silos contain: which functions sit in the same org unit as the function that creates legal obligations, and which functions are systematically separated from it. In regulated industries across domains, this pattern is consistent: the functions that might discover adverse findings about the entity's products or operations are systematically separated from the functions that would create legal obligations if they received those findings.
To understand the Liability Partition, it is necessary to understand how "knowledge" operates as a legal category. In the law of products liability, negligence, and regulatory compliance, the question is not what individual employees knew, but what the institution knew. Institutional knowledge is attributed to a corporation through a set of legal rules that determine when an employee's knowledge binds the corporation. In general, corporate knowledge is attributed when the employee who had knowledge was acting within the scope of their authority, when the knowledge was relevant to a matter within that authority, and when the employee either had or should have had the knowledge in the exercise of their authority.
The institutional consequence of these rules is specific: the corporation knows what its decision-makers know in the exercise of their decision-making authority. A quality control technician's knowledge of a defect on the line is not institutional knowledge in the relevant sense if the technician has no authority over the decision that generated the defect and no channel by which the knowledge is expected to reach the decision-maker. A safety engineer's knowledge of a design flaw is not institutional knowledge if the engineer is not in a reporting chain that reaches product approval authority. The silo that separates safety engineering from product approval is, simultaneously, the legal architecture that ensures the safety engineer's knowledge does not constitute institutional knowledge of the flaw.
If the Liability Partition theory is correct, it predicts a specific pattern in which functions get siloed from which other functions, and that prediction is testable against documented organizational structures in regulated industries. The prediction: functions that generate potentially adverse findings about institutional products, operations, or safety will be systematically separated — organizationally, geographically, or through reporting structures — from the executive functions whose knowledge would create legal obligations.
The prediction holds across documented cases. In pharmaceutical companies, clinical safety monitoring functions are typically separated from commercial operations — the safety group reports findings to the medical affairs function rather than to the commercial team that makes decisions about product promotion and sales targets. In the rare cases where these functions are integrated — where safety monitoring informs commercial strategy in real time — the commercial team's awareness of safety signals creates legal exposure on the commercial decisions made after those signals were received. The silo protects the commercial function from that exposure.
In platform companies, trust and safety has historically been separated from product and revenue — the team that evaluates content welfare impacts does not report to the team that sets engagement optimization targets. This separation means that the product team's decisions about feed algorithm design are not made with institutional knowledge of the welfare findings that the trust and safety team has generated. Frances Haugen's disclosure in 2021 revealed that Facebook's internal research had generated findings about Instagram's effects on adolescent mental health that were not shared with the product team — not because they were suppressed through malice, but because the organizational structure did not route them there. The silo did what silos do.
| Industry | The function generating adverse findings | The function whose knowledge creates legal obligations | The silo between them | Legal effect of the separation |
|---|---|---|---|---|
| Pharmaceutical | Pharmacovigilance / post-market safety monitoring | Commercial / product promotion decision-makers | Medical affairs versus commercial separation; safety findings routed through regulatory, not commercial, channel | Commercial decisions about promotion made without institutional knowledge of safety signals — limiting liability for promotion of products with known adverse events |
| Aviation | Safety engineering / FMEA (failure mode and effects analysis) | Program management / certification authority | Safety analysis documentation separated from certification submissions; safety engineers without authority over certification decisions | Certification authority's approval decisions made with formal record that safety analysis is complete, without institutional knowledge of open safety findings in engineering documentation |
| Financial services | Internal audit / model risk management | Trading desk / product structuring decision-makers | Risk management reports to risk committee, not to trading desk; audit findings routed to compliance, not to deal teams | Trading decisions made without institutional knowledge of model limitations identified by model risk team; deal structuring decisions made without institutional knowledge of audit findings on similar structures |
| Food manufacturing | Quality assurance / allergen testing | Production management / scheduling decision-makers | QA reports to corporate quality, not to plant management; production scheduling decisions made by plant operations | Production scheduling decisions — including decisions about line changeover procedures — made without institutional knowledge of QA findings about allergen cross-contact risks on the specific line |
| Platform governance | Trust and safety / internal welfare research | Product / algorithm design decision-makers | Trust and safety reports to legal or policy, not to product; welfare research conducted by separate team without product authority | Product decisions about engagement optimization made without institutional knowledge of welfare findings — maintaining that product team decisions were made in good faith without awareness of harm evidence |
The most direct evidence for the legal function of organizational siloing comes from the involvement of legal counsel in organizational design decisions — specifically, decisions about which functions should report to which other functions and how findings should be routed through the organization. In the tobacco case, internal documents show that the structure of research programs — including which researchers reported to which functions and how findings were handled — was influenced by legal counsel's assessment of what organizational arrangements would protect the company from institutional knowledge liability. The Shook Hardy structure, in which the industry's primary outside law firm managed the flow of research findings, was an organizational design decision with explicit legal function.
More generally, in regulated industries where litigation risk is high and well-understood — pharmaceutical, financial services, aviation — there is a documented practice of involving legal counsel in organizational design not only for legal functions but for the operational functions that might generate findings with legal consequences. The presence of legal counsel in discussions about which functions report to which, and how adverse findings are routed, is a signature of organizational design being used for liability management rather than (or in addition to) coordination efficiency.
The development and commercialization of tetraethyl lead as a gasoline additive by Standard Oil of New Jersey and General Motors in the 1920s provides one of the most fully documented cases of a Liability Partition in industrial history. The case is documented in primary sources — congressional testimony, corporate archives, and the extensive research record developed by Clair Patterson and others decades later — and illustrates the Liability Partition in a period before the legal doctrine had been fully articulated.
The key factual structure: by 1923, there was documented evidence within the manufacturing operations that tetraethyl lead posed serious health risks to production workers. The Deepwater, New Jersey plant — operated by Standard Oil's Ethyl Corporation subsidiary — saw workers experiencing lead poisoning at rates that were undeniable; several died, and dozens suffered neurological damage visible enough to become public knowledge. The public manufacturing crisis was managed by a combination of statements minimizing the occupational hazard and assertions that street-level gasoline exposure was too dilute to cause harm.
The organizational structure that managed the liability was specific: the operational knowledge of worker harm was contained within the manufacturing subsidiary (Ethyl Corporation), which was separate from the parent corporations (Standard Oil, General Motors). The public health questions about broader population exposure from automobile exhaust were routed to the Public Health Service for study — a study the industry funded and whose scope it influenced. The research function (such as it was) was separated from the commercial function. The individuals with operational knowledge of worker harm were production employees of the subsidiary, not executives of the parent companies whose knowledge would have created commercial liability for the product.
The result: tetraethyl lead was added to gasoline at commercial scale for fifty years after the initial evidence of harm, removing it only after independent research — funded without industry involvement — documented the population-level blood lead burden and its cognitive consequences. The organizational separation between operational harm knowledge and commercial decision-making authority held for half a century. This is the Liability Partition at scale.
The Liability Partition explains the organizational structure of the Accountability Firewall. But the silo alone does not explain why the individuals within the siloed function — the quality engineers, the safety analysts, the trust and safety researchers — do not act unilaterally on what they know. Those individuals face a specific structural trap: they know, they want to act, and the organizational structure makes effective action individually catastrophic while making inaction collectively sustainable. AF-002 examines this trap — the Treading Lightly Problem — as the human face of the Accountability Firewall.
Organizational siloing is a genuine coordination mechanism with well-documented efficiency benefits. The separation of safety functions from commercial functions is often mandated by regulation precisely to prevent commercial pressure from distorting safety decisions. Interpreting every silo as a liability strategy imputes malicious intent to what are often neutral organizational design choices.
The objection is well-founded on its own terms. The Liability Partition does not require malicious intent and does not deny that siloing serves legitimate coordination functions. The point is not that siloing is always or usually a liability strategy — it is that in regulated industries with known litigation risk, the legal function of organizational structure is evaluated explicitly and shapes design decisions in ways that produce a predictable pattern: the functions that discover adverse findings are separated from the functions whose knowledge creates obligations. This pattern is consistent across industries and over time. Intent may vary; the structural effect does not.
Internal: This paper is part of Accountability Firewall (AF series), Saga VI. It draws on and contributes to the argument documented across 23 papers in 5 series.