Public AI Governance Documentation

NIST AI RMF Alignment

How HireProxy maps current controls and next steps to Govern, Map, Measure, and Manage.

Last reviewed: July 6, 2026

NIST describes the AI RMF Core as four functions: Govern, Map, Measure, and Manage. This document maps HireProxy's current controls and reasonable next steps to those functions. It is not a certification or external audit.

Govern

NIST intent: Establish accountability, policies, roles, risk culture, documentation, incident processes, and third-party risk management.

Current HireProxy controls:

• Broadlake Technologies LLC is the accountable operator.
• Public privacy, data retention, terms, AI risk assessment, and incident response pages describe platform-level governance.
• Model identifiers are centralized so model upgrades can be reviewed in one place.
• Terms prohibit fabricated credentials, impersonation, unauthorized voice recording, and undisclosed live-interview assistance.
• Candidate sites include an "About this assistant" disclosure that AI can make mistakes and final details should be confirmed with the candidate.
• Core candidate-provided profile data can be exported, edited, unpublished, or deleted.
• Sensitive product surfaces are excluded from session diagnostics.

Reasonable next controls:

• Maintain this AI governance package as the public control narrative.
• Keep a quarterly AI risk review log.
• Add an AI system inventory owner field and review date for each system card.
• Maintain the vendor/model register before provider or model upgrades.
• Keep an incident register for AI quality, privacy, security, provider, and voice-practice incidents.

Map

NIST intent: Understand the system context, intended use, users, benefits, risks, affected parties, data context, and deployment environment.

Current HireProxy controls:

• Intended use is limited to candidate-owned career representation, role-fit analysis, interview preparation, and practice.
• Public terms state that HireProxy is not an automated employment selection, ranking, or rejection system.
• Sensitive or higher-risk employment decisions are explicitly reserved for humans outside the product.
• Public and private data boundaries are documented: career-site content can be public, while raw interview practice answers are private by default.
• Interview Mode maps context by company, role, interview stage, interviewer context, saved prep, transcript, and delivery metrics.
• Public career agents are scoped to one tenant's candidate data.

Reasonable next controls:

• Add a short pre-release risk review for any new AI surface.
• Keep system cards updated when an AI surface, model, provider, or data flow changes.
• For future B2B or outplacement sales, document customer-specific context, administrator responsibilities, and any additional human review expectations.

Measure

NIST intent: Evaluate, test, monitor, and document AI risks, including trustworthiness, privacy, security, transparency, fairness, and reliability.

Current HireProxy controls:

• Automated tests cover accuracy guardrails, privacy telemetry controls, interview prep, public discovery, and rehearsal behavior.
• Error monitoring captures application exceptions and provider failures.
• Usage events track interview prep and rehearsal completion.
• Report-abuse intake allows external users to flag misleading or false candidate-site content.
• Voice features enforce ownership checks, rate limits, entitlement checks, and input size limits.
• The voice feedback architecture preserves full detailed scoring while improving responsiveness.

Reasonable next controls:

• Build a small grounding/confabulation eval set with representative public career-agent questions, fit memo prompts, and interview feedback transcripts.
• Track a simple AI quality metric: confirmed AI-content issues per 100 meaningful AI interactions.
• Track report-abuse categories: false fact, privacy issue, inappropriate content, prompt injection, impersonation, and other.
• Add a release checklist item for AI changes: AI instruction changed, model changed, provider changed, data flow changed, public disclosure changed, retention changed, evaluation run attached.
• Review voice transcription edge cases quarterly if voice usage increases.

Manage

NIST intent: Prioritize, respond to, reduce, transfer, monitor, or accept risks. Communicate incidents and errors. Deactivate or roll back systems when needed.

Current HireProxy controls:

• Incident response page defines AI content, prompt injection, cross-tenant data leakage, service disruption, and voice/practice data incident types.
• Candidate assistants can be unpublished or disabled at a tenant level.
• Platform-wide AI surfaces can be restricted by deployment changes or provider configuration.
• Candidate edits refresh downstream AI context used for future responses.
• Account deletion removes user data, cancels subscription where possible, and deletes the auth user.
• Public data retention explains deletion and retention boundaries.

Reasonable next controls:

• Use the incident register template for every confirmed AI quality incident.
• Define severity levels: - Sev 1: confirmed cross-tenant data exposure, security breach, or personal data exposure. - Sev 2: materially false AI statement likely to affect hiring, reputation, or user trust. - Sev 3: low-impact incorrect or poor-quality AI output. - Sev 4: harmless quality issue or documentation gap.
• Maintain standard response targets appropriate for a solopreneur: - Sev 1: contain immediately, notify affected users when confirmed, document root cause and corrective action. - Sev 2: investigate within one business day, correct content or AI instructions, and notify the affected candidate when appropriate. - Sev 3 and Sev 4: batch into weekly or quarterly quality review unless patterns emerge.
• Revisit risk posture before selling to enterprises, universities, outplacement firms, or other institutional customers.