Louis Frettman
Most ChatGPT data policies are written as if every employee in the organization represents identical risk. The acceptable use policy prohibits submitting company data to unauthorized AI tools. The network filter blocks ChatGPT for everyone. The training session covers the policy with the full staff simultaneously. The controls are uniform because designing uniform controls is simpler than designing controls calibrated to actual risk — and because the organization has not yet worked through what “actual risk” means when mapped to specific roles and the data those roles access.
The uniformity creates two simultaneous problems that pull in opposite directions. Over-restriction for lower-risk employees — the marketing coordinator whose AI use involves public brand information, the operations staff member whose AI tasks involve internal scheduling and logistics — generates adoption friction and resentment that produces workarounds. Under-protection for higher-risk employees — the CFO who uses AI to analyze financial models, the HR director who uses AI to draft employee communications, the attorney or legal coordinator who uses AI to review contracts — creates data exposure where the consequences of a breach are most significant.
A role-based approach addresses both problems simultaneously. It identifies which roles carry elevated data exposure risk based on the sensitivity and strategic value of the data those roles routinely handle, applies controls proportionate to that risk level, and allows the sanctioned alternative to carry the AI productivity function for all roles — which is how the access restriction avoids creating the adoption problem that unrestricted blanket policies attempt to avoid by not restricting at all. This article describes the role-based risk identification process, what calibrated controls look like at different risk levels, and how to manage the role-based framework as the organization evolves.
Identifying Your High-Risk Roles — Where Company Data Exposure Is Most Consequential
The starting point of a role-based AI data risk assessment is identifying which roles in the organization routinely access data categories where AI-related exposure would be most consequential — either because the data is regulated, because it constitutes trade secret or competitive intelligence, or because its disclosure would create significant liability or business harm. Three role categories consistently surface as elevated risk across small and mid-size business environments.
Finance and Executive Leadership — Strategic and Financial Data
Roles with access to financial performance data, financial models, pricing strategies, budget and forecast information, and any merger, acquisition, or investment-related information represent the highest-risk category in most small business environments. The data these roles handle is simultaneously the most competitively sensitive and the most valuable to external actors who might benefit from its disclosure.
CFOs and finance staff using ChatGPT to analyze financial models, draft investor communications, review contract terms, or prepare board reporting materials are submitting some of the organization’s most sensitive strategic information through consumer AI channels. The analysis help they are seeking is legitimate and the productivity benefit is real — which is precisely why the control architecture for this role category needs to provide a compliant alternative rather than simply blocking access and expecting the work to be done without AI assistance.
Executive leadership — owners, CEOs, presidents, and senior partners — represents an elevated risk category for a specific reason beyond the data they access: they are frequently exempt from the organizational controls that apply to staff. Executives who believe that AI data policies apply to employees but not to themselves, or who are not included in the access restriction implementation because IT staff are hesitant to apply controls to senior leadership, represent the highest-consequence uncontrolled risk in the organization. An executive using a personal ChatGPT account to process strategic planning documents, financial projections, M&A discussion materials, or board presentation content is generating an exposure that no staff-level control framework addresses. Executive inclusion in role-based AI data governance is not optional — it is the element most commonly missing from implementations that are otherwise well-designed.
HR and Legal — Employee Data and Privileged Communications
Human resources roles handle data categories that are both legally sensitive and personally significant in ways that create specific liability exposure when they enter unsanctioned AI channels. Employee compensation data, performance review documentation, disciplinary records, leave and accommodation records, and the communications associated with employee relations matters are all categories that HR staff use AI to assist with — drafting performance improvement plans, preparing termination documentation, summarizing employee history for management discussions, and generating standard HR communications.
The sensitivity of this data is distinct from financial sensitivity. Compensation data in a consumer AI tool creates pay equity and confidentiality exposure. Disciplinary documentation creates employment litigation risk if it can be characterized as having been processed through channels that do not protect its confidentiality. Accommodation records may constitute protected health information under applicable frameworks. HR roles require AI governance controls calibrated to the specific legal landscape of employment data rather than the general framework applied to other data categories.
Legal and quasi-legal roles — in-house counsel, paralegals, compliance officers, and contract managers — handle information that may be subject to attorney-client privilege, work product protection, and professional responsibility obligations. Legal information processed through consumer AI tools enters those tools under terms that do not preserve privilege, potentially constituting a privilege waiver that affects the organization’s ability to assert privilege in subsequent litigation. This is not a speculative risk — it is a recognized concern that bar associations have addressed in AI guidance, and it represents a specific and significant consequence of unsanctioned AI use in legal functions.
Sales and Business Development — Client and Competitive Intelligence
Sales and business development roles represent elevated risk because they handle two categories of high-value information simultaneously: client data that is typically subject to confidentiality provisions in client agreements, and competitive intelligence — pricing strategies, proposal approaches, competitive win/loss analysis, pipeline information — that constitutes the organization’s primary competitive asset.
Sales professionals using ChatGPT to draft proposals frequently submit client-specific information, proposed pricing, competitive context, and the organization’s service positioning — a comprehensive package of competitive intelligence in a single AI interaction. Business development managers using ChatGPT to analyze pipeline data and identify priority opportunities are submitting the organization’s full business development strategy. Account managers using ChatGPT to prepare for client meetings may submit relationship history, client financial information, and service delivery details that are covered by confidentiality provisions in the client contract.
The client confidentiality dimension of sales role AI use creates contractual exposure that can be triggered by a client security questionnaire, a client contract review, or a client incident — events that occur at unpredictable times and that may surface ChatGPT use that occurred months or years before the triggering event.
Calibrating Controls to Role Risk Level
Role-based AI data governance applies controls proportionate to the risk level identified in the role assessment. The control set for a high-risk role is more restrictive and more technically enforced than the control set for a standard-risk role — but both role categories receive a sanctioned alternative that ensures AI productivity support remains available through an approved channel.
High-Risk Role Controls — Stricter Technical and Policy Measures
High-risk roles — finance, executive leadership, HR, legal, and sales at the account management level — require a control set that goes beyond policy acknowledgment and network-level blocking. The characteristics of high-risk data, and the professional sophistication of the roles that handle it, make technical controls that depend on network position insufficient. An executive working from home on a personal device is outside the corporate network. A sales professional on a client site is using cellular data. A legal coordinator using their personal phone to draft communications is invisible to endpoint monitoring.
High-risk role controls should include device-level management for company-issued devices that enforces AI tool restrictions regardless of network position, explicit personal device use policies for work purposes that address AI tool use specifically, role-specific policy training that covers the specific data categories and specific AI risks relevant to the role rather than a general AI policy, and a sanctioned AI environment configured to handle the high-sensitivity data categories the role routinely processes — with the data handling agreements, access controls, and audit logging that those data categories require.
For executive leadership, the control framework should be presented as an enablement rather than a restriction — the executive receives AI capability through the sanctioned environment that is more capable, more secure, and more appropriate for strategic-level work than the consumer alternatives, rather than receiving an instruction that AI tools they have been using are now restricted.
Standard-Risk Role Controls — A Balanced Approach That Supports Adoption
Standard-risk roles — marketing, operations, administrative, customer service, and similar functions whose primary AI use involves less sensitive data categories — require a control set that governs without generating the friction that produces workarounds. For these roles, the policy-and-alternative approach described in most AI acceptable use frameworks is appropriate: a clear policy about what data categories may not enter unsanctioned AI tools, a sanctioned alternative that handles the productivity tasks these roles need AI for, and network-level restrictions on consumer AI tools that are visible enough to reinforce the policy without creating the adversarial relationship that aggressive technical enforcement can produce.
The calibration for standard-risk roles should recognize that these roles’ AI productivity needs are genuine and legitimate, and that the control architecture’s job is to meet those needs through the sanctioned channel rather than to restrict AI use as an end in itself. A marketing coordinator who cannot get AI assistance for content drafting through the organization’s sanctioned environment will find a way to get it through an unsanctioned one. The sanctioned environment’s coverage of standard-risk use cases is the control that matters most for this role category.
The Sanctioned Alternative Requirement Across All Risk Levels
The role-based risk framework only functions if every role category has a sanctioned AI alternative that handles their legitimate AI productivity needs. This is the structural requirement that makes role-based controls sustainable — without it, the restriction creates adoption failure at every risk level, because every employee has legitimate AI productivity needs regardless of their risk profile.
The sanctioned alternative for high-risk roles needs to be capable of handling sensitive data under appropriate governance: enterprise data handling agreements, role-based access controls, and audit logging. The sanctioned alternative for standard-risk roles needs to cover the productivity use cases those roles rely on most heavily. A sanctioned environment that handles the use cases of one risk tier but not another will produce compliant behavior in the covered tier and workarounds in the uncovered one, defeating the role calibration that the framework was designed to create.
Managing the Role-Based Framework as the Organization Evolves
A role-based AI data control framework is not a static configuration. Roles change when employees are promoted, when responsibilities are restructured, when new functions are created, and when organizational priorities shift. Each of these changes potentially affects an employee’s risk tier — a promotion from sales coordinator to sales director may move an employee from standard to high risk, a restructuring that gives an operations manager budget oversight may add financial data to their risk profile, a new business development function may create a role category the framework did not originally address.
Maintaining the framework requires connecting AI access tier assignments to the HR and IT processes that manage role changes. When an employee’s role changes, their AI access tier should be reviewed and updated as a standard component of the role change process — not as a separate IT security review that happens inconsistently, but as a defined step in the role transition workflow that occurs automatically.
Regular review of the role tier assignments themselves is also required as the organization’s AI use cases develop. A role that was standard-risk when the framework was built may become high-risk as that function’s AI use expands into data categories the initial assessment did not anticipate. Annual review of tier assignments against actual role AI use, rather than only against role definitions, ensures the framework remains calibrated to actual risk rather than to the risk profile that existed at the time of initial implementation.
Building the infrastructure to genuinely prevent employees from using ChatGPT with company data requires a governance architecture calibrated to the real risk landscape of the organization — one that recognizes where risk is highest, applies controls proportionate to that risk, and sustains those controls through the organizational changes that affect role risk profiles over time.
The NIST AI Risk Management Framework addresses role-based accountability and access governance within its Govern function — providing a structured approach to defining which organizational roles bear which AI-related responsibilities and how access to AI systems should be managed in alignment with data sensitivity and organizational risk tolerance.
The NIST Special Publication 800-53 security and privacy controls framework addresses role-based access control as a formal security control category — providing the technical and policy framework for implementing access controls that vary by role based on the sensitivity of the data and systems the role is authorized to access, which is directly applicable to AI system access governance in organizational environments.