Key takeaways
- Vet an AI coach like any HR system, then go further. Ask for SOC 2 Type 2 and ISO 27001, plus clear answers on how the AI uses your data.
- Get data use in writing. Your data should never train AI models, and conversations should be private by default.
- Scope access everywhere. Single sign-on with MFA, role-based admin rights and field-level control over HR data.
- Treat each integration as a data flow. Workday, Teams, your identity provider and MCP connections each need a scope and a log.
- Govern what the AI does. Log every AI action, keep people approving decisions about people, and agree incident and exit terms.
What should IT check before rolling out an AI coach?
Before rolling out an AI coach, IT should check thirteen things: certifications, data residency, model training on your data, access controls, conversation privacy, audit logs, SSO and provisioning, HRIS integration, collaboration tools, AI assistant and MCP access, the vendor’s AI governance, incident response and exit terms. Most are standard for any HR system. What changes is the content: employees tell a coach about their manager, their performance and their doubts, so data use and privacy weigh more.
Use the table as your questionnaire, and ask to see each answer in the product.
| Check | What to ask | Why it matters | A good answer |
|---|---|---|---|
| Certifications | Which independent audits do you hold? Can we see the reports? | Third-party audits test controls, not promises. | A current SOC 2 Type 2 report and ISO 27001 certificate; FedRAMP for U.S. public sector. |
| Data residency | Where is our data hosted? Can we choose the region? | Privacy laws and works councils may require data to stay in a region. | Regions you choose, a data processing agreement (DPA) and a published subprocessor list. |
| Model training | Is our data, including conversations, used to train AI models? | Training could carry employees’ words into a model others use. | No, in the contract, for the vendor and its AI model providers. |
| Access controls and admin scoping | Who at our company and at the vendor can see which data? | Admin roles decide who can reach employee data. | Role-based access, admins scoped to their programs, vendor access reviewed. |
| Conversation privacy | Can managers, HR or admins read coaching conversations? | If employees think their manager can read along, they stop being candid. | Private by default; employees decide what to share; HR sees aggregates only. |
| Audit and logging | What is logged, including actions the AI takes? | You need to reconstruct who or what touched data, and why. | All access and user activity, plus a record of every AI action. |
| SSO and provisioning | Do you support SAML or OIDC single sign-on, MFA and automated provisioning? | Access should end the day someone leaves. | SSO through your identity provider, MFA enforced there, SCIM or HRIS-driven provisioning. |
| HRIS integration | Which HRIS fields do you read, and can we exclude some? | The HRIS also holds compensation and other sensitive data. | A field list you approve, with sensitive fields left out. |
| Collaboration tools | What does the app read and post in Teams, Slack, email and calendars? | Every app approved in your tenant widens where data flows. | Least-privilege permissions and a documented list of what it reads and posts. |
| AI assistants and MCP | Can AI assistants or our own agents reach the coach? How is that secured? | An MCP connection lets another AI read from, or act in, the coach. | A documented MCP server or API, per-user sign-in, admin control and logging. |
| Vendor AI governance | How do you manage AI risk? Which features inform decisions about people? | The EU AI Act treats some AI uses in employment as high-risk. | A recognized framework, human approval for decisions about people, reasons shown. |
| Incident response | What is your incident process, and when will you notify us? | A breach here exposes very personal content. | A formal, tested process, a notification window in the contract, regular penetration tests. |
| Exit and data deletion | What happens to our data when the contract ends? | Coaching history shouldn’t outlive the relationship. | A stated retention policy, export on request and confirmed deletion. |
Is AI coaching secure enough for enterprise use?
Yes, when the vendor meets the bar you already set for HR systems and adds controls for the AI itself. The usual bar is independent audits, encryption, single sign-on and role-based access. The AI layer adds no training on your data, private conversations, scoped access to HR data and a log of everything the AI does.
An enterprise AI coaching platform deserves extra scrutiny because:
- The content is personal. Employees talk to a coach about a tense relationship with a manager, a bad review or whether to stay.
- It reads from your systems. Coaching draws on role, team and goals, so the vendor holds organizational data too.
- The AI may act. AI agents that send nudges or propose mentor matches need permissions, limits and approvals, like any service account.
- It reaches into other tools. Chat apps, calendars and AI assistants extend where data flows.
Red flags: no SOC 2 Type 2 report or ISO 27001 certificate; a vague answer on model training; admins who can open individual conversations; no list of subprocessors, including AI model providers; no record of what the AI did. New to the category? Read what an AI coach is and how it works.
Which AI coaching platforms let us control how employee data is used?
Choose a platform that gives you control at four points: which HR data the coach can read, what employees share and with whom, what the vendor may do with the data, and how long it is kept. Rumi by Chronus, for example, lets administrators choose which organizational data Rumi Coach can access, never uses data they exclude, keeps employee conversations private by default and never uses your data to train AI models.
Controls to confirm in a demo:
- Field-level HR data scope. You pick the fields the coach can use and leave out compensation or anything else.
- Employee control over sharing. Employees see anything before it goes to a manager or mentor, and can take it back.
- Aggregate-only reporting. Participation and progress across groups, never one person’s conversations.
- No training, in the contract. Get it in the DPA, not only on a website.
- Subprocessors and retention. A published list of who processes your data, including AI model providers, and a stated retention policy.
What AI governance should HR software have?
AI governance for HR software comes down to four things: clear rules for what the AI may read and do, people approving decisions about people, a record of every AI action, and a vendor that manages AI risk against a recognized framework.
Three references help HR, IT and legal review it together:
- The NIST AI Risk Management Framework, released in January 2023, is a voluntary framework for managing AI risks to individuals, organizations and society.
- ISO/IEC 42001, published in December 2023, sets requirements for an AI management system in organizations that provide or use AI.
- Under the EU AI Act, which entered into force in 2024, the European Commission lists AI tools for employment and management of workers among high-risk uses. Rules for those uses apply from December 2, 2027.
Questions for the vendor:
- Which features make or inform decisions about people, such as promotions, task allocation or performance reviews?
- Can we require a person’s approval before the AI takes a step, such as proposing a mentor match?
- Does each AI suggestion show its reasons?
- Is every AI action logged: what ran, when, why and which records it touched?
Which AI coaching tools support MCP integration?
AI coaching tools that publish an MCP server support MCP integration: AI assistants and agents your company already uses can connect to the coach through a standard protocol. Rumi by Chronus, for example, offers an MCP server for any MCP-compatible assistant and an SDK to embed it in your portals and tools, with agent-to-agent handoff on the roadmap.
The Model Context Protocol (MCP) is an open-source standard for connecting AI applications to external systems. For AI coaching it means two things: employees can reach coaching from the AI assistant they already use, and your own AI agents can hand work to the coach.
An MCP connection is another door into employee data, so review it like an API:
- What it exposes. Which tools and data the server offers, and whether they only read or can also write.
- Who is acting. Each call should run as a signed-in user with that user’s permissions, not as a shared account.
- Which assistants connect. Admins should approve the AI clients that can connect.
- What is logged. Every call recorded, like any other access.
- Whose terms apply. What employees type into an assistant also falls under its own data terms.
How does an AI coach connect to Workday, Teams and Okta?
An AI coach connects to each one at a different layer. Workday, or another HRIS, gives the coach context such as role, team and role changes, usually through a vendor-managed connector or API. Microsoft Teams is where employees get notifications, meeting invites and reminders. Okta, or another identity provider, handles sign-in through SAML or OpenID Connect (OIDC) single sign-on and, where the vendor supports it, account provisioning through SCIM.
| System | What flows | What to check |
|---|---|---|
| HRIS (Workday) | Role, manager, team, location, and events such as hires, promotions and role changes | Which fields sync and how often; whether you can leave out compensation and other sensitive fields |
| Collaboration (Microsoft Teams) | Notifications, reminders, meeting links and calendar events | Permissions requested in your tenant, what the app reads and posts, admin approval |
| Identity (Okta or another provider) | Sign-in, MFA, and account creation and removal | SAML 2.0 or OIDC, MFA enforced at the identity provider, SCIM provisioning and deprovisioning |
SCIM (System for Cross-domain Identity Management) is an open IETF protocol, published in 2015, for managing user identities across systems. With it, accounts are created when people join and removed when they leave.
The same checks apply to mentoring software integrations, which read the same HR data and book meetings in the same calendars.
How Rumi by Chronus answers the checklist
Rumi Coach is the AI coach in Rumi by Chronus, which brings AI coaching, mentoring and communities into one platform. Here is how it answers each point.
| Checklist point | Rumi by Chronus |
|---|---|
| Certifications | SOC 2 Type 2, ISO 27001 and CSA STAR certified. FedRAMP Moderate Authorized, listed as FedRAMP Certified on the FedRAMP Marketplace since April 2025, with a DoD Impact Level 4 provisional authorization. |
| Data residency | Hosted on AWS in U.S. East, Europe or Australia, with a government cloud option. GDPR compliant, with a data processing agreement and a published subprocessor list. |
| Model training | Your data is never used to train AI models. |
| Access controls | Role-based access control across product and infrastructure; access to Chronus systems is granted by role and reviewed quarterly. Administrators choose which organizational data Rumi Coach can access, and any data they exclude is never used. Each AI agent reads only the fields its job needs. |
| Conversation privacy | Private by default. Employees choose what to share with a manager or mentor, see it before it’s sent and can take it back. Your talent team sees participation and progress in aggregate, never anyone’s private conversations. |
| Audit and logging | Logs of all access and user activity, centralized in a SIEM platform and monitored for anomalies. Every agent action is logged: what ran, when, why and which records it touched. |
| SSO | Single sign-on through SAML and other mechanisms. Multi-factor authentication can be enabled through SSO. |
| HRIS | Rumi connects to Workday, SAP SuccessFactors and your other HR systems. You connect your systems once, and every Rumi product shares them. Sync the fields you pick; leave out compensation and other sensitive fields. |
| Collaboration tools | Rumi Mentor works with tools you already use, like Workday, Slack, Teams, Zoom and Outlook. People sign in once, and meetings show up on their calendar. Chronus also supports Google Meet, Google Calendar, Apple Calendar, Microsoft Exchange and Microsoft 365. |
| AI assistants and MCP | Employees can ask Rumi Coach for help in Slack, ChatGPT or Claude. For IT and AI teams, Rumi offers an MCP server for any MCP-compatible assistant and an SDK, with agent-to-agent handoff on the roadmap. |
| AI governance | Orchestration sets the rules, agents do the work and people make the calls. Every agent has an owner, a defined scope and a full action log; you choose which actions wait for sign-off; every recommendation shows its reasons. |
| Incident response | Formal incident response policies and procedures, including for data breaches; regular automated and manual penetration tests; disaster recovery simulations every six months; cyber insurance. |
| Retention | Chronus retains customer data only as long as its data retention policies or applicable laws require. Backups are encrypted with AES-256. Details are in the Chronus Trust Portal. |
Approvals, guardrails and data scope are configuration, not code. See how orchestration records every run, how Rumi’s AI agents work within scoped access, and all integrations.
Review Chronus security and compliance
Frequently asked questions
- Does an AI coach need its own security review?
Yes. An AI coach adds new data flows: personal conversations, HR context and connections to chat tools and AI assistants. Add AI-specific checks, such as model training, conversation privacy, AI action logs and MCP access, to your standard questionnaire.
- Can managers or HR read employees’ AI coaching conversations?
They shouldn’t. In a well-designed AI coach, conversations are private by default, employees decide what to share, and HR sees participation and progress only in aggregate. With Rumi Coach, employees see anything before it’s sent to a manager or mentor and can take it back.
- Do AI coaching vendors train AI models on customer data?
Policies differ, so ask, and get the answer in the contract. Check the vendor’s AI model providers too, through its subprocessor list. With Rumi by Chronus, your data is never used to train AI models.
- What certifications should an AI coaching platform have?
At minimum, a current SOC 2 Type 2 report and ISO 27001 certification; U.S. federal agencies also look for FedRAMP authorization. Chronus is SOC 2 Type 2, ISO 27001 and CSA STAR certified, FedRAMP Moderate Authorized and holds a DoD Impact Level 4 provisional authorization.
- Who should own an AI coaching rollout, HR or IT?
Both. HR designs the program and owns the coaching experience; IT connects systems and governs identity, data scope and security. In Rumi, HR and program teams build and edit playbooks themselves, and IT connects systems once.
