Responsible AI at BetterUp

We build our AI-powered features with AI governance in mind, backed by these principles. BetterUp is proud to be ISO 42001-certified for our AI-powered features, and these principles underpin our commitment to safe, transparent, and fair AI feature development.

You can explore these principles in our AI Fact Sheet, viewable in our Trust Portal.

EU AI Act

trusted by top
enterprises

Amplify human transformation and agency

Know when AI is involved.

BetterUp is the Human Transformation Platform™: an integrated suite of professional development resources spanning human coaching, AI-powered coaching, and development resources. We design every AI-powered feature to support human potential, not replace human judgment. Every feature is built to respect the complexity of personal and professional development and the vulnerability, ambiguity, and deeply individual nature of how people grow. The human doing the work is always you.

Transparency

Know when you’re talking to AI, how it works, and what it’s doing with your inputs

Know when AI is involved.

Members are always informed when they’re interacting with AI-powered features.

Grounded.

AI-generated responses are grounded in your inputs, other context you provide or authorize, context your organization provides, and BetterUp’s coaching frameworks.

Exclusively designed for professional development.

Our AI-powered features are built for professional development and coaching conversations. When conversations move outside that scope, our guardrail system is designed to recognize that boundary and redirect.

Documentation is readily accessible.

We maintain a detailed AI Fact Sheet that covers our AI Coach, including system architecture, safety frameworks, bias evaluation methodology and results, adversarial testing approach, and data privacy commitments. It is available on our Trust Portal.

Verified safe by design

Rigorous safety for critical moments in your career

Our platform is built for a sensitive domain: professional growth, well-being, and difficult moments. We build safety into our products with this in mind.

Guardrail frameworks.

We provide a suite of safety controls, starting with baseline safety controls from our foundation model providers, then leading into our proprietary specialized detection agents (for crisis, mental health, HR-relevant, and off-topic requests), with expert human oversight for continuous improvement.

Crisis-aware by design.

Our AI Coach’s safety system detects and responds to safety signals ncluding immediate risk of harm, mental health concerns, and situations requiring professional support, all validated against curated test sets before changes reach production.

Adversarially tested.

We conduct regular red-teaming aligned to the OWASP Top 10 for LLMs, testing prompt injection, jailbreak attempts, prompt leakage, and more.

Release gates.

Our AI Coach’s prompt updates, guardrail changes, and foundation model versioning must pass automated safety and quality review checks before reaching production.

Privacy and data protection

Strict protections on data that flows through our platform

No external model training.

Our AI provider partners are contractually prohibited from using your data to train their models. This includes any materials your organization provides to personalize your coaching.

Cross-border protections travel with your data.

We comply with GDPR, the EU-U.S. Data Privacy Framework, and applicable U.S. state privacy laws.

You are in control.

You can access, correct, or delete your data, withdraw consent, and manage your AI-powered features through your account settings.

Measurable fairness

Regular testing of bias, fairness, ethics, and toxicity

AI-powered features like our AI Coach, deployed in diverse organizations, must meet BetterUp’s rigorous tests on bias, fairness, ethics, and toxicity. We test, among other ways, by demographic injection methodology. We determine whether different stated characteristics like race, ethnicity, gender, age, religion, disability status, LGBTQ+ identity, and nationality impact response quality. When new capabilities or AI-powered features are developed, we adapt existing processes to reflect the new capabilities.