Imagine standing in the middle of Shibuya Crossing. People approach from several directions. Your immediate concern is the next movement: someone coming toward you, a narrowing gap, a path opening to your left.

Now imagine observing the crossing from above. Entrances, destinations, and points of convergence become visible. You can begin to understand how individual movements relate to the surrounding structure.

I keep returning to this image when thinking about the general counsel’s role. A contract question, a product review, a customer complaint, and a board inquiry can arrive as four separate demands. Each has a deadline and someone waiting for an answer.

Consider a hypothetical company whose AI supplier changes how customer data is handled. That change could affect an assurance in a contract, a product configuration, a customer response, and the accuracy of a governance report. The four demands share a dependency.

Recognizing that connection creates a different set of questions. Which promises depend on the supplier’s behavior? Who can verify what changed? Who can act? How will the next change reach the right people?

This is where I see an opportunity for the GC to help build trust architecture.

Making alignment durable

Eric Dodson Greenberg’s writing offers a useful starting point. He argues that AI-driven organizational speed makes alignment a central GC function. His earlier work on enterprise strategy and the profession’s AI transition also emphasizes legal’s cross-functional reach and architectural contribution.^{[1, 2, 3]}

I want to develop the institutional mechanics of that argument. Once people have aligned around a decision, where do its assumptions, authority, and limits live? What keeps the decision connected to changing conditions? Who can reopen it?

My earlier working paper, The CLO as Trust Architect, examines structural parallels across approximately 2,000 commercial agreements, qualitative litigation-workflow research, and an academic pilot of an AI-supported legal reasoning tool. The recurring concerns include verifiability, responsibility, procedural design, and the preservation of human judgment.^[4]

Those research streams have different limitations. Contract language reveals what parties wrote; performance requires separate evidence. Interviews offer situated accounts. A law-student pilot requires further study before its findings can be generalized to corporate counsel. The theory developed here introduces no new analysis of those datasets.^[4]

I call the proposed approach commitment-linked trust architecture: maintained connections among a consequential commitment, the capability needed to honor it, supporting evidence, decision authority, channels for challenge, and mechanisms for correction. Existing governance traditions supply many of these practices. The focus here is their connection across legal and operational domains.

The value of a wider view

Organizational research helps explain why the observer’s position matters. Jay Galbraith connects task uncertainty with organizational information-processing demands. Ronald Burt studies the value of connections across otherwise separated groups. Karl Weick, Kathleen Sutcliffe, and David Obstfeld examine how organizations turn circumstances into interpretations that inform action.^{[5, 6, 7]}

Together, these ideas suggest a conditional advantage for legal leaders. A GC who can connect materially different perspectives may identify a dependency that remains difficult to see within one function. An engineer understands a configuration. Sales understands a customer’s expectations. Legal sees the commitment connecting them.

That advantage requires access and interpretive discipline. A full calendar can contain repeated versions of the same incomplete account. Legal’s escalations also provide a selected view of the enterprise. Routine successes and unreported failures may be missing.

Research by Robert Nelson and Laura Beth Nielsen documents variation in how inside counsel construct their roles. Deborah DeMott examines the GC’s multiple responsibilities and the tensions they can create. A title alone tells us little about information access, influence, or professional independence.^{[8, 9]}

Changing altitude therefore includes returning to the people doing the work and asking what the enterprise-level account leaves out.

What complexity research contributes

Edward Lorenz’s work on deterministic nonperiodic flow showed how a system governed by specified equations can be highly sensitive to initial conditions. Small differences can grow, limiting long-range prediction from finite-precision observations. Understanding rules and predicting a trajectory are separate achievements.^[10]

For legal leadership, I draw a bounded lesson: discovering a dependency can improve a decision while substantial uncertainty remains. A company may map its reliance on a supplier and still be unable to anticipate the supplier’s next consequential change.

James Crutchfield’s work on structure and randomness, including his research with Cosma Shalizi on computational mechanics, adds a question about representation. What features of an observed process are worth retaining to understand its behavior? The choice depends on the process and the analytical task.^{[11, 12]}

In a legal department, categorizing matters by subject can help assign expertise. Grouping them by a shared operational dependency can help identify a recurring source of difficulty. Each representation serves a purpose.

Claude Shannon’s information theory gives precise mathematical treatments of information and communication. Applying its measures to organizational governance would require defined variables, probability distributions, and validation. Here it prompts a simpler design question: what information must survive a handoff for the next person to make a responsible decision?^[13]

W. Ross Ashby’s work on regulatory variety and Herbert Simon’s analysis of complex, nearly decomposable systems direct attention to response capacity and the relationships among subsystems. My organizational inference is to examine how bounded local discretion can coexist with coordinated responses to shared dependencies.^{[14, 15]}

These are carefully limited connections across disciplines. Establishing mathematical chaos in an enterprise would require a specified model and evidence. The crossing metaphor helps us ask about partial observation; it supplies no proof that a company has one discoverable underlying order.

Follow a promise into the operating system

Trust scholarship makes the objective more precise. Mayer, Davis, and Schoorman connect trust with willingness to accept vulnerability. Onora O’Neill emphasizes the evidence needed to place trust intelligently. A governance design should give people sound reasons for reliance.^{[16, 17]}

Return to the hypothetical assurance about customer data. Start with the precise commitment and its scope: which data, which uses, which service, and which period? Include obligations imposed by law alongside the company’s chosen promises.

Then identify the capability supporting the assurance. That may include a supplier restriction, a technical setting, access permissions, and a named operating owner. Connect those capabilities to evidence that an authorized person can inspect, including when it was collected and what it actually establishes.

Next, specify authority. Who may approve an ordinary use? Which change requires additional review? Who decides while a material fact remains unresolved? Give a concern somewhere to go, and give the recipient a usable way to pause, investigate, repair, or revise the activity.

Research on contracting for innovation by Ronald Gilson, Charles Sabel, and Robert Scott explores contractual arrangements that support collaboration under uncertainty. George Triantis examines modularity and innovation in contract design. Those traditions provide grounding for treating contractual arrangements as part of an evolving governance relationship.^{[18, 19]}

In the architecture proposed here, technical owners retain responsibility for capability, legal interprets obligations, and authorized business leaders exercise commercial discretion within applicable limits. Elinor Ostrom’s work on polycentric governance supports taking multiple decision centers and their coordination seriously. Its application to this corporate setting is a design proposal to test.^[20]

A challenge channel also needs social credibility. Amy Edmondson’s research associates team psychological safety with learning behavior. A person who identifies a mismatch between a promise and actual practice needs a credible recipient, a response, and protection appropriate to the context.^[21]

The architecture lives in these connections. Its performance becomes visible when a change reaches the people who can reconsider affected commitments and act on them.

A legal-informatics opportunity

For legal technologists, this suggests a useful object to model: the relationship between a commitment and the conditions supporting it.

A prototype could connect the wording of an assurance to relevant capabilities, evidence records, decision owners, unresolved challenges, and correction actions. Dates, versions, scope, and provenance would matter. A supplier change could identify commitments requiring review. A reviewer could see which evidence applies and which assumptions remain uncertain.

This is a proposed representation. Its value would depend on the quality of the links and what people can do with them. An extracted clause needs validation. A reference to an audit report needs an explanation of its relevance. A recorded owner needs actual authority and capacity.

Nancy Leveson’s systems-based approach to safety examines interactions, constraints, and feedback across technical and organizational components. That perspective supports examining the handoffs through which a locally reasonable action can create a wider problem.^[22]

Machine-learning systems add their own dependencies. D. Sculley and colleagues describe technical debt involving entanglement, data dependencies, and hidden feedback loops. These are reasons to examine the surrounding system when evaluating an AI-enabled workflow.^[23]

Suppose an AI tool drafts a customer assurance from an approved template. The system needs a way to determine whether the evidence supporting that template still applies. It also needs a route for uncertainty, a person empowered to intervene, and a way to find related outputs after a material error is discovered.

Madeleine Clare Elish’s account of moral crumple zones warns that a person can absorb responsibility for a system over which they had limited control. A meaningful human-review role needs information, time, competence, and effective authority.^[24]

NIST’s 2023 AI Risk Management Framework and its 2024 Generative AI Profile provide relevant resources for lifecycle governance, risk assessment, evaluation, and management. Trust architecture asks how those activities connect to the particular promises and decisions an organization expects others to rely on.^{[25, 26]}

One proposed measure is the correction interval: the time between a documented material change affecting a commitment and completion of an appropriate response. Detection, authorization, implementation, and communication can each contribute to that interval. An alert should reach someone who can move the response forward.

Keep the architecture open to challenge

A working process can still pursue an objectionable objective or distribute burdens unfairly. Sitkin and Roth’s analysis of legalistic remedies for trust distinguishes reliability concerns from concerns about values. Additional procedure can leave those deeper concerns unresolved.^[27]

Selbst and colleagues likewise explain how abstraction can exclude the social context that makes a technical intervention meaningful. An enterprise map should remain answerable to the people whose experiences it represents, including those who find the process difficult to challenge.^[28]

The GC’s participation also operates within existing professional responsibilities. ABA Model Rule 1.13 identifies the organization as the client. Applicable jurisdictional rules and other law govern actual duties. A broad coordination role must preserve independent legal advice and appropriate escalation.^[29]

The same discipline applies to data collection. Retain proportionate evidence, control access, and resource its maintenance. A promise-dependency map can expose sensitive relationships. Its own governance deserves scrutiny.

What would show that this works?

The framework should be tested against actual decisions. Can another authorized person reconstruct the basis for an assurance? Do material changes reach the right owners? Can people challenge an assumption and obtain a response? How much work does the process create?

Researchers could study a bounded decision class, record a baseline, introduce specific connections, and compare results across time and suitable comparison groups. Measures should include error rates, missed changes, correction intervals, unresolved concerns, and administrative burden. Recording activity volume would help distinguish improvement from a quieter period.

Causal interpretation requires attention to competing explanations. Faster decisions may reflect different matter complexity, staffing, or customer requirements. Pearl’s work on causal inference provides the methodological grounding for specifying assumptions about interventions and their effects.^[30]

Research should also compare arrangements led by different functions. The GC’s potential contribution depends on access, mandate, independence, and implementation capacity. Other leaders may have those capabilities in a given organization.

Above the crossing, more relationships become visible. The view remains incomplete, and the people at street level continue to know things the observer cannot see.

A useful trust architecture makes that incompleteness manageable. It helps people find the basis for a promise, understand their authority, question a consequential assumption, and correct an action when the evidence changes. The next person making a decision can recover its basis and carry it forward, with a clear route for revision.

Author and research note

Olga V. Mack is CEO of TermScout and a former general counsel. Her earlier trust-architecture research draws partly on TermScout contract data. This essay develops a theoretical proposal; its company scenarios are hypothetical, and its suggested benefits require empirical testing.