NIST separates GenAI risk evidence from coaching outcome evidence

Classify each claim before accepting its evidence

A coaching-system presentation may combine security, responsible AI, coaching quality, participant experience, behavior change, sponsor value, and business impact in one narrative. Separate them before diligence begins. For every claim, identify the subject, population, configured use, evidence type, measurement period, comparison, owner, and limitation. NIST's Generative AI Profile is a cross-sector companion to AI RMF 1.0 concerned with trustworthiness across the design, development, use, and evaluation of AI systems. That makes it relevant to GenAI risk work, but it does not transform a provider's alignment statement into evidence that coaching is effective.

Use two linked but distinct records. The system-risk record asks how the configured product identifies, measures, manages, and governs relevant GenAI risks. The coaching-evidence record asks whether the service, method, and participant experience support the specific leadership objective and what outcomes were observed under a defined method. Connect the records where system behavior can affect coaching, such as fabricated guidance, unsafe escalation, confidentiality loss, or inappropriate personalization. Do not collapse them: a control can reduce a risk without producing a leadership outcome, and a positive participant report can coexist with unexamined technical risk.

Scope GenAI risk evidence to the configured service

Record the product version, model and provider roles, prompts or other governing instructions at a functional level, retrieval sources, memory, data flow, integrations, human review, participant and sponsor access, and actions the system can take. Then ask the provider to map its NIST-related claims to the actual configuration the buyer will use, including evidence dates, tests, exceptions, residual risks, responsible owners, and material changes. A generic policy, framework logo, or enterprise-wide statement does not show how an executive-coaching conversation behaves with the selected model, account, data, administrator settings, and escalation path.

Test representative and adverse situations at the service boundary: inaccurate or invented content, conflicting goals, sensitive disclosure, manipulation, bias, over-reliance, prompt or data leakage, unsafe advice, inaccessible interaction, and failure of the human handoff. Preserve the expected response and what actually happened. Evaluate whether the platform communicates uncertainty, keeps the participant able to stop or correct the interaction, and prevents the sponsor from receiving information outside the agreement. NIST calls the AI RMF voluntary; using its language does not mean NIST assessed, approved, or certified the product.

Build coaching outcome evidence around the decision

Define the coaching claim independently of the technology claim. Name the participant population, leadership job, baseline, service dose, coach and AI roles, outcome measure, observation window, response rate, comparison where appropriate, and important conditions such as employer sponsorship or prior experience. Distinguish activity measures—logins, sessions, prompts, completion, satisfaction—from evidence of learning, behavior, decision quality, role performance, or organizational results. If the provider reports an improvement percentage, request the denominator, method, exclusions, uncertainty, and source data rather than assuming a polished dashboard supplies them.

Include participant agency, confidentiality, accessibility, coach or professional boundaries, sponsor effects, escalation, and unintended consequences in the outcome review. A system can increase engagement while narrowing reflection, increasing dependence, or exposing sensitive material. A human coach can also shape results that should not be credited entirely to the AI feature. Preserve qualitative accounts as accounts and label editorial inference. NIST AI 600-1 did not study coaching populations or validate a coaching method, so it cannot supply the missing causal, comparative, or longitudinal evidence.

Reopen both records when the system changes

A new model, memory feature, retrieval source, sponsor dashboard, risk classifier, or automated action can change both the GenAI risk profile and the coaching experience. Define change triggers before approval: what requires renewed testing, participant notice, coach preparation, contract review, data assessment, outcome-baseline revision, or a pause. Keep versioned evidence so an evaluation of the prior service is not represented as current after a material release. Assign separate accountable owners for technical risk and coaching evidence, with one decision forum able to see where the records interact.

At renewal, report supported conclusions, unresolved questions, incidents, exceptions, population limits, and evidence age without averaging unlike claims into a single trust score. The NIST profile supplies a voluntary risk-management resource for GenAI across sectors; it does not establish legal compliance, clinical safety, professional coaching quality, accessibility, confidentiality, product fit, or outcomes. Procurement should therefore require both records to be strong enough for the proposed use and should refuse the shortcut in either direction: responsible-AI language is not outcome evidence, and an outcome story is not risk governance.

Turn this source into a reviewable decision

For AI Coaching Platforms for Leadership Development, use this briefing as a dated decision record rather than a substitute for the source. Preserve National Institute of Standards and Technology, the exact URL, the July 27, 2026 review date, the supported facts above, the editorial interpretation, the limitations, and any buyer-specific evidence. Link that record to the decisions most directly affected: Validation and outcome evidence; Coaching method and content provenance; Data flow and confidentiality; Safety, boundaries, and escalation. State whether the source changes the scope, evidence requirement, control, sequence, or only the language used to describe the decision.

Before action, name the accountable owner, affected population and workflow, exact offering or configuration, source data and rights, human decision point, exception and appeal path, complete cost, expected benefit, failure and stop conditions, retained evidence, and next review date. Keep official facts, provider statements, buyer observations, representative tests, measured outcomes, editorial inferences, and unknowns visibly separate. Reopen the record when the source, offer, model, integration, data, policy, population, responsible person, or measured result changes.

Limitations and unknowns

NIST AI 600-1 is a voluntary, cross-sector Generative AI Profile and companion to AI RMF 1.0. It is not a coaching standard, product assessment, certification, legal or clinical determination, or study of coaching effectiveness. Applying it to a particular system requires current configuration, risk, participant, professional, organizational, and outcome evidence.