Protecting Confidentiality in AI-Assisted Leadership Coaching

Why confidentiality is the first design decision

The main risk in AI-assisted coaching is not only accidental disclosure. It is the gradual normalisation of oversharing. When a leader uses a chatbot to draft a reflection, rehearse a difficult conversation, or summarise a session, the tool may retain content, route it through third-party systems, or make it available under terms the user has not read closely. MIT Sloan’s guidance on broadly applicable generative AI tools notes risks such as data loss, intellectual property exposure and security breaches, which is why sanctioned access and clear usage rules matter ³.

From a coaching perspective, confidentiality should shape the workflow before any tool is chosen. A practical rule is simple: if a prompt would not be comfortable in a meeting room with the wrong people present, it should not be entered into a public or uncertain AI system. That applies to names, team issues, performance concerns, board matters, health information and any identifying detail that could reveal the leader, their colleagues or the organisation.

Separate reflection from record

Harvard Business School’s work on leading AI teams suggests that effective leaders ask more questions, create turn-taking and adapt on the basis of what they hear ². That insight is useful here because good coaching uses AI for preparation, not confession. A leader can ask a model to help frame options, test assumptions or draft neutral questions without feeding in the full story. The reflection stays with the leader; the AI receives only what is necessary.

A useful safeguard is to split material into three levels. Keep the personal interpretation, emotions and sensitive context offline. Share only the smallest factual slice needed for a specific task. If an AI output is useful, transfer the insight back into the coaching conversation rather than storing the original prompt as part of the formal record. This reduces the chance that a later audit, breach or vendor change exposes more than intended.

Set controls that coaches and leaders can actually follow

University guidance on generative AI governance consistently favours cross-functional rules rather than ad hoc individual judgement ³⁴. For coaching programmes, that means privacy, legal, information security and the coaching lead should agree the permitted tools, approved data types and retention expectations. The policy should also make clear whether prompts and outputs may be saved, shared with managers, or used in programme evaluation. Without that clarity, even well-meaning coaches can create inconsistent practice.

In day-to-day use, three habits do most of the work. First, use the least revealing prompt possible. Second, prefer organisation-approved tools with known data handling terms. Third, keep a clean separation between coaching notes, AI drafts and formal HR records. If a leader needs to refer to a difficult interpersonal issue, it is safer to discuss the matter in supervision or live coaching than to ask a system to analyse it from a detailed transcript. The coaching value lies in the thinking, not in building a searchable dossier.

Treat AI as a drafting partner, not a vault

Stanford’s AI Index shows rapid adoption alongside rising concern about responsible AI staying behind capability ¹. That pattern matters for coaching because the temptation is to use AI as a convenient memory store. It can summarise themes, suggest next questions or reword a communication plan, but it is not a confidential archive by default. Organisations should be careful about using consumer tools as long-term repositories for coaching material.

A better model is to use AI briefly and selectively. Ask it to help structure a difficult conversation, compare decision options or identify likely blind spots, then delete the session if the platform and policy allow. Where a tool offers enterprise controls, those controls still need to be checked against the organisation’s privacy expectations. The goal is not to ban AI from coaching; it is to avoid letting convenience outrun consent.

An Authenticate advisory perspective

Our practical view is that AI can add value to leadership coaching when it is used as a bounded thinking aid, with the human coach holding the confidentiality line. If the boundaries are unclear, the benefit quickly falls away because trust is the medium through which coaching works.