RESPONSIBLE AI
Built on clear principles. Not just compliance.
CoachBot's approach to AI in coaching is grounded in four non-negotiable commitments — and a governance structure that keeps them current.
Our commitments
Four non-negotiable commitments. Maintained by design, not policy.
No AI model fine-tuning on client data.
Sensitive client and organizational data stays out of model training entirely — no exceptions. Your conversations never become training data.
No human-like avatars, voice cloning, or face cloning.
The research on anthropomorphism risks in digital coaching contexts is clear. We will not cross this line regardless of demand or competitive pressure.
No psychometric or personality profiling assessments.
High-risk AI use cases exist under the EU AI Act for good reason. We do not build them, and we do not integrate tools that do.
No engagement-loop design.
CoachBot conversations have a defined start and end. We optimize for client outcomes — not session frequency, platform retention, or time-on-app metrics.
Responsible AI Coaching Quality Council
Our council actively monitors emerging research in AI coaching, anthropomorphism, and LLM governance — ensuring these commitments stay current, not just compliant.
THE PROBLEM
AI coaching needs a shared governance model. Most platforms don't offer one.
With most AI coaching tools, the vendor controls the system prompt and the organization has no visibility or input. There is no shared governance — just vendor lock-in.
THE COACHBOT MODEL
Shared responsibility. Clear accountability at every layer.
CoachBot owns the infrastructure layer — guardrails, compliance, and security that cannot be switched off. You own the coaching layer — your methodology, your brand, your quality standards. Accountability is clear at every level.
Governance layer — always on, never negotiable.
Guardrails
System prompt & custom moderation
AI security
External security review: an independent technical audit, completed July 2026. Security controls continuously monitored in Vanta.
Layered memory
User administers their own settings
Models
Gemini via Vertex AI in the EU, the default, included in every plan · On Scale: ChatGPT (OpenAI), Claude (Anthropic) or Mistral with your own API key (BYOK)
Certification
SOC 2 Type II observation period started October 2026 (report expected Q1 2027); ISO 27001 to follow.
EU hosted
GDPR · EU AI Act · No model fine-tuning on client data
THE RISK
Employees won't use tools they don't trust.
If employees suspect their conversations are monitored or shared without consent, they disengage. The coaching value disappears — and so does the ROI.
THE COACHBOT APPROACH
The employee is in control. Always.
Every coachee decides what their coach or HR can see. Nothing is shared without explicit consent. Trust isn't a feature — it's the foundation the whole system is built on.
Built for enterprise procurement. Let's talk.
Enterprise procurement and legal teams can request additional documentation or book a governance conversation with our team.