AI Governance Statement

The StaffApp™
Effective Date: April 17, 2026  |  Last Updated: April 17, 2026  |  Next Review: April 2027

1. Purpose and Scope

This AI Governance Statement describes how Chef Life Media LLC deploys, monitors, and is accountable for the artificial intelligence components of The StaffApp™ (the “Service”). It applies to every AI-driven feature in the Service, both customer-facing and internal. It is written for customers, end users, procurement and security teams, and regulators who want to understand our posture before, during, and after an engagement.

This Statement supplements (and does not replace) our Terms of Service, Privacy Policy, Acceptable Use Policy, and Data Processing Agreement. Where specific obligations conflict, those documents control; this Statement provides context and commitments beyond the minimum legal floor.

2. Our AI Principles

The StaffApp™ was built on the conviction that hospitality leadership is a human craft and that technology should strengthen that craft, not replace it. Our AI principles follow from that belief:

1. AI augments manager judgment. It never replaces it. Every AI-generated suggestion is advisory. Consequential employment actions — hiring, discipline, termination, promotion, pay changes — must be decided by a human manager, reviewed against the facts of the situation, and documented in the manager’s own words.
2. We prefer transparency over capability. When a choice exists between a more capable AI feature that hides how it works and a less capable one that is explainable, we choose explainable. Customers are entitled to know what the AI sees, what it does with what it sees, and why it suggested what it suggested.
3. Customer Content is not training data. Your staff records, check-in notes, discipline documentation, DevCoach conversations, and voice inputs are never used to train our AI models or any of our subprocessors’ models. This is contractual with our subprocessors, not just a policy preference.
4. AI should reduce bias in management decisions, not introduce it. We actively monitor AI-involved text for language that indicates potential bias, retaliation, or protected-characteristic references, and we surface those to managers for review. Our goal is to help managers catch their own blind spots, not to create new ones.
5. We document the coaching system behind the AI, and we name its source. DevCoach’s frameworks, progressive discipline guidance, and leadership language are derived from the Chef Life coaching system developed by Chef Adam M. Lamb — not from scraped internet data or unattributed corpora. Attribution is public and verifiable at cheflifecoaching.com and thecheflifebrigade.com.
6. Human reviewers read what the AI writes before it becomes a decision. DevCoach suggestions are reviewable, editable, and discardable. A manager who accepts a DevCoach suggestion verbatim is still the author of record for the resulting Staff Record entry, and is accountable for it as if they had written it themselves.

3. Where AI Is Used in The StaffApp™

The following is the complete current inventory of AI-involved features in the Service. We maintain this list here as the canonical map; any change is reflected here first.

We are not using AI for any other purpose in the Service today. There is no AI-driven resume screening, no AI-driven scheduling, no AI-driven performance scoring or ranking, and no AI-assisted generation of disciplinary language or termination recommendations. If that changes, this page changes first, and customers are notified under Section 8.

4. Human Oversight and Decision-Making

The StaffApp™ is not an automated decision-making system within the meaning of applicable AI laws (including the EU AI Act, NYC Local Law 144, the Colorado AI Act, and similar frameworks). Every consequential action involving a staff member is a human decision by a human manager.

4.1 What DevCoach Does and Does Not Decide

DevCoach generates suggestions. It does not:

• Take any action in a Staff Record without a human manager’s explicit confirmation
• Submit, finalize, or publish check-ins, incidents, or discipline documentation autonomously
• Send notifications or communications to staff members without human approval
• Score, rank, or rate individual employees
• Recommend a specific hire, termination, or promotion decision

4.2 Progressive Discipline Framework

Where DevCoach references a progressive discipline framework in its coaching suggestions, it operates from the industry-standard sequence of one verbal warning, followed by up to three written warnings, prior to a termination decision. This framework is advisory guidance reflecting general best practice; it is not a legal standard and it does not override the customer’s own disciplinary policies, collective bargaining agreements, or applicable employment law. Customers are expected to follow the discipline procedures appropriate to their operation, jurisdiction, and workforce.

Customers may upload or configure their own knowledge base or specific procedures into the Service to override or supplement the default framework. Customer-provided guidance takes precedence over Chef Life Media’s default coaching content when in conflict.

4.3 The Manager Is the Author of Record

When a manager accepts a DevCoach suggestion — verbatim, edited, or as a starting point — the resulting Staff Record entry is the manager’s own statement for all legal and employment purposes. The manager is accountable for its accuracy, appropriateness, and compliance with applicable law.

5. Training Data and Model Sourcing

5.1 Foundation Models We Use

We use third-party foundation models for the generative AI components of the Service. We do not train our own foundation models. Our vendors and their model provenance are:

• Anthropic (Claude Haiku): A commercial large language model trained by Anthropic. Anthropic publishes model cards and responsible-scaling commitments at anthropic.com.
• Voyage AI (text embeddings): A commercial embedding model used to convert text into vectors for retrieval. Voyage does not receive or store Customer Content; embeddings are computed on our curated knowledge base content only.
• Deepgram (speech-to-text): Commercial speech recognition model operating under a zero-retention configuration for our API requests.
• ElevenLabs (text-to-speech): Commercial voice synthesis, using a custom voice model trained by ElevenLabs from licensed voice samples of Chef Adam M. Lamb. The voice model is a derivative work licensed for use in the Service.
• DeepL (translation): Commercial neural translation model. Translation is initiated only on explicit user action.

5.2 Our Knowledge Base

DevCoach’s coaching content is grounded in a proprietary knowledge base curated by Chef Life Media LLC. All source material is derived from the Chef Life coaching system, specifically:

• Transcripts and notes from actual coaching calls conducted by Chef Adam M. Lamb
• Published books authored by Chef Lamb
• Articles written by Chef Lamb and published through Chef Life Media properties
• Podcast episode transcripts from Chef Lamb’s hospitality leadership podcast catalog

No scraped internet content, no anonymized customer conversations, and no unattributed third-party corpora are included in the DevCoach knowledge base. All content in the knowledge base has been personally vetted by Chef Lamb.

5.3 What We Do Not Do

• We do not use Customer Content to train any AI model, ours or any subprocessor’s.
• We do not fine-tune foundation models on Customer Content.
• We do not permit our subprocessors to use API-submitted data for training under their applicable commercial terms.

6. Bias and Fairness Monitoring

6.1 Today: Trigger-Word and Phrase Monitoring

The Service maintains a curated list of trigger words and phrases that indicate potentially biased, discriminatory, retaliatory, or safety-concerning language. Any written text entered into a check-in, incident report, discipline record, or other Staff Record field is scanned against this list at the time of entry. When a match is found, the record is flagged and a notification is generated to the relevant leadership role for review before the record is finalized in the normal workflow.

This mechanism is designed to help managers catch their own blind spots — language about a protected characteristic that may not belong in a disciplinary record, language that sounds retaliatory, language that might indicate a wellness or safety concern. It is a safety net for equality and equity, not a substitute for manager training or judgment.

6.2 What This Is Not

The trigger-word monitoring system is pattern matching, not an AI classifier. It does not score employees, it does not score managers, and it does not make any employment decision. It surfaces text for human review.

6.3 Commitment to Formal Bias Testing

We commit to implementing a formal, documented bias-testing regimen for the DevCoach generative AI components of the Service by July 1, 2026. This will include, at minimum:

• Defined test scenarios covering protected characteristics and employment contexts
• Sampling of DevCoach outputs against those scenarios
• Manual review of flagged outputs by trained reviewers
• Documented findings with remediation plans for any patterns identified
• Annual re-testing with updated scenarios

Until that formal testing program is in place, we rely on the combination of (a) our upstream model providers’ own bias testing and safety work, (b) our trigger-word monitoring layer described in Section 6.1, and (c) the human-review architecture described in Section 4. We believe this layered approach is reasonable for the current stage of the Service, but we are not satisfied with it as a long-term posture, which is why formal testing is on the calendar.

7. User Rights Related to AI

As a user or an end user of the Service, you have the following AI-related rights:

• Right to know that AI was used. Any DevCoach output is labeled as AI-generated within the Service. Voice output via the Listen feature is clearly identified as a synthesized rendering.
• Right to opt out of AI features. Customer Organization administrators may disable DevCoach, voice dictation, and voice output for individual users or the whole organization through the in-app settings.
• Right to human review. Any employment-affecting action must be taken by a human manager. If you are an employee whose record you believe was influenced by an AI suggestion, you may request human review by contacting your Customer Organization’s administrator; escalation paths are described in our Privacy Policy.
• Right to information about the coaching system. The Chef Life coaching system that grounds DevCoach is publicly documented. See cheflifecoaching.com and thecheflifebrigade.com.
• Right to request explanations. If you believe an AI-involved feature produced an output you cannot understand or that seems inappropriate, contact legal@thestaffdevapp.com with a description and we will review.

8. Transparency and Change Management

Material changes to our AI architecture — adding a new AI feature, changing a foundation model, engaging a new AI subprocessor, or meaningfully changing the purpose of an existing AI feature — trigger the following commitments:

• This page is updated before or at the same time as the change takes effect in the Service.
• If the change involves a new subprocessor, our Subprocessor List is updated with at least 30 days’ advance notice per our Data Processing Agreement.
• If the change affects data handling in a way that alters our Privacy Policy, Customer Organizations receive email notice at least 30 days in advance.
• If the change materially reduces customer control or privacy protections, affected customers have the right to cancel without penalty within the notice period.

9. Incident and Misuse Handling

9.1 When AI Produces a Bad Output

AI models can produce inaccurate, inappropriate, biased, or otherwise bad outputs. When you encounter one in The StaffApp™, please report it to legal@thestaffdevapp.com with the feature, the output, and enough context for us to reproduce the problem. We review every report and respond within five business days.

We treat repeated patterns of problematic output as a governance issue, not a support ticket. Patterns that indicate bias, safety failures, or policy violations are escalated to the founder and may result in model changes, prompt changes, feature restrictions, or vendor changes.

9.2 When Users Try to Misuse AI Features

Our Acceptable Use Policy prohibits using DevCoach to harass, harm, deceive, or generate content that would violate employment law. It also prohibits prompt injection, jailbreaking, and role-play framings designed to bypass safety guidelines. Violations are grounds for suspension or termination under the AUP.

9.3 When a Vendor Has an Incident

If one of our AI subprocessors experiences a security incident, model regression, or policy violation that materially affects customers, we notify affected Customer Organizations in accordance with our Data Processing Agreement breach notification timelines (within 72 hours of confirmed impact).

10. Regulatory Alignment

The StaffApp™ is not currently classified as a high-risk AI system under any major AI governance framework, because we do not use AI for automated employment decisions, creditworthiness determinations, biometric categorization, or other use cases those frameworks regulate. We nonetheless align voluntarily with relevant elements of the following frameworks:

• EU AI Act. We disclose AI use, maintain this governance documentation, ensure human oversight, and use a text-to-speech subprocessor (ElevenLabs) that holds ISO/IEC 42001:2023 certification for AI management systems.
• NYC Local Law 144 (Automated Employment Decision Tools). We do not operate as an AEDT within the meaning of the law because no consequential employment decision is automated. Customer Organizations using the Service in New York City remain responsible for their own AEDT compliance if they deploy automated tools elsewhere in their hiring or promotion workflow.
• Colorado AI Act. We do not make or assist in making consequential decisions as defined by the Act. Customers operating in Colorado should review their own compliance obligations.
• California AI laws (SB 1001, AB 331, and related). We support customer compliance by providing transparency disclosures, human-review controls, and AI-feature opt-outs. Customers remain responsible for their own jurisdiction-specific notices to employees.
• White House AI Bill of Rights framework and NIST AI RMF. Our principles in Section 2 and practices in Sections 4, 6, and 7 are consistent with the safe-and-effective-systems, algorithmic-discrimination-protections, data-privacy, notice-and-explanation, and human-alternatives principles articulated in those frameworks.

Customers with specific regulatory obligations should contact legal@thestaffdevapp.com to discuss documentation needs, customer-specific Data Processing Agreements, or other compliance artifacts.

11. Review and Updates

This Statement is reviewed at least annually, with the next scheduled review noted at the top of this page. It is also updated within 30 days of any material change to the AI architecture or governance posture described above. Significant updates are summarized in a changelog maintained internally and available on request to legal@thestaffdevapp.com.

Non-material updates (typo corrections, link updates, clarifying edits) may be made without advance notice but are reflected in the “Last Updated” date.

12. Contact

Questions about our AI governance posture, requests for additional documentation for a procurement or security review, or reports of problematic AI outputs should be directed to: