PI AI security & trust pack

This article applies to all organizations using Obi, either in PI2 or through the Predictive Index MCP. (Requires an active PI Inspire subscription.)

Version 1.1 — Last Updated October 2026

What Obi and the MCP are, and what data they use
Common Obi and MCP questions and answers
How your data flows through Obi
PI anonymizes data before and model training
Who can access your Obi data
Where your data lives, and for how long
Security certifications and testing
Responsible AI and employment law
Controls you have
The agreements that apply
Current limitations
Questions, contacts, and further reading

1. What Obi and the MCP are, and what data they use

Obi by The Predictive Index

Obi by PI delivers behavioral intelligence: guidance grounded in 70 years of behavioral science and your organization’s own PI data. A manager types a question in plain language — “How should I open a tough feedback conversation with Sam?” — and Obi answers using the behavioral data of the actual people involved.

Obi works with 4 kinds of data:

Your PI behavioral data: Assessment results, job targets, and team data already in your PI account; Obi sees only what the person asking can already see.

Prompts: The questions people type, which may mention employees by name.

Company context (optional): Your admin can add a website link, strategy text, and up to 3 .pdf, .txt, or .docx files. Everyone in your account can see it; only admins can change it.

Outputs: The guidance that Obi generates.

Currently, access to Obi requires an active PI Inspire subscription and works alongside PI Inspire, Hire, Design, and Diagnose. Your admin turns it on or off for the whole organization — using it is always optional.

Predictive Index MCP

The Predictive Index MCP uses the Model Context Protocol (MCP) — an open standard — to bring Obi into the AI tools your team already uses: Claude, Microsoft Copilot, and ChatGPT. Once enabled, users can ask behavioral intelligence questions from inside their AI host and get answers grounded in PI’s validated behavioral science, with citations. The connector is read-only and stateless; the AI host sends the question to Obi and relays Obi’s validated answer back to the user.

Use of the PI MCP requires access to Obi and thus an active PI Inspire subscription.

2. Common Obi and MCP questions and answers

Question	Answer
Is our data used to train third-party AI models?	Obi in PI2: No, your data is not used to train any third-party models. Obi via PI MCP: Not within PI's pipeline. Inside your AI host, your organization's agreement with that host governs training. (See: Section 4.)
Does PI use our data to train its own private model?	Yes — but PI only uses your anonymized data to train its own private model. All names and any other information that would identify your people or company is removed. The only way to opt-out of this use, is to turn Obi off.
Which AI provider does Obi use?	PI utilizes a private model built on OpenAI (ChatGPT) technology. Providers may change over time.
Can we turn Obi off?	Yes. Your admin can enable/disable Obi for the whole organization at any time, without affecting other PI products.
Who can see our people’s Obi prompts?	Each conversation in Obi is private to the person having it. No one else at your company — not managers, not Org Admins — can see it. A small PI product/engineering team can access prompt data, only to monitor and improve quality and security. That access is governed under PI's SOC 2 Type 2 and ISO 27001 certified controls, and it's never used to evaluate individual employees.
Is our data kept separate from other customers’ data?	Yes. Data is logically separated; training is never pooled across accounts; access is zero-trust.
Where is data processed and stored?	Microsoft Azure, eastern U.S. Encrypted in transit (TLS 1.3+ enforced) and at rest (AES-256).
How long is data kept?	Client data is anonymized 120 days after the end of the contract, unless requested earlier by the client.
What certifications does PI hold?	ISO 27001:2022, SOC 2 (Type 2 report available), PCI-DSS 4.0, NIST CSF alignment; GDPR compliant. Such certifications include PI's AI features as applicable.
Is there a list of subprocessors?	Yes. The list of subprocessors is available here.
Is there a Data Processing Agreement?	Yes. Details can be found here.
Does Obi make hiring or firing decisions?	No. Obi is trained not to make decisions even when asked, and PI monitors that guardrail. Review AEDT definitions with your own counsel as they vary from state to state.
Who owns the prompts and answers?	You own your prompts. Answers are mixed: You own your data inside them; PI owns its content inside them.
Is there a data flow diagram?	Yes — see the illustrative diagram in the next section. The Trust Center provides a network diagram on request.

3. How your data flows through Obi

Tip: Visit PI’s Trust Center for a more detailed flow.

4. PI anonymizes data before and model training

PI always removes all identifying information from your data before it is used to train or improve our own private model. This means the data can no longer be connected to an individual or company.

Your data is not used to train any third-party model that our products are built on. We have contracts in place preventing third-parties from using our clients’ data to train their models.

PI uses your anonymized data to improve its own private model. We are continuously monitoring the prompts and outputs to ensure our private model is operating as intended and abiding by the safety guardrails we have established.

Does the PI MCP server train AI models on our people data?

Inside PI's pipeline, we never train third-party models on your data, and we only use anonymized data to improve our own products. Inside the AI host you connect to, that host's agreement with your organization governs how data is retained or used — including for training. Enterprise tiers of the major AI platforms each publish commitments around customer data; we recommend reviewing yours.

5. Who can access your Obi data

Your own people, per your own rules. Role-based access control with 4 user roles plus a dedicated third-party consultant role. Data segregation restricts access by org unit or geography.

Obi follows those same rules. It never shows anyone information they couldn’t already see in PI. All existing permissions are honored.

Conversations are private. A person’s Obi chats are not visible to their manager, admin, or executives.

Inside PI, access is need-to-know. Support and engineering access stored data only when troubleshooting; DevOps for deployments; a small subset of our product/engineering team accesses prompt data only to check and improve quality.

Sign-in security. OAuth 2.0 through Auth0, multi-factor authentication, automated protection against brute-force login attacks. Learn more about SSO.

6. Where your data lives, and for how long

Location: Microsoft Azure, eastern United States. All PI subprocessors are U.S.-based (list located here).

Encryption: In-transit production systems enforce TLS 1.3 or above; at rest, AES-256.

Resilience: In-scope databases replicated across regions at least 300 miles apart.

Retention: Logs kept ≥12 rolling months. After contract end, client data is anonymized at 120 days — earlier upon request.

EU data: Covered by PI’s DPA with Standard Contractual Clauses.

7. Security certifications and testing

Certifications and frameworks: ISO 27001:2022 certified, SOC 2 audited (Type 2 report available), PCI-DSS 4.0 validated, NIST Cybersecurity Framework alignment, GDPR compliant. Such certifications include PI's AI features as applicable.

Penetration testing: Annually, by third-party security firms. Findings and fixes are verified during SOC 2 and ISO audits; an executive summary is available in the Trust Center.

Documents on request via the Trust Center: ISO certificate, SOC 2 Type 2 report, PCI validation, pen-test executive summary, network diagram, information security policy, certificate of insurance. Access is granted to existing customers and prospective customers may access the Trust Center by executing an NDA at trust.predictiveindex.com.

Live status: Available at status.predictiveindex.com.

Found a vulnerability? Responsible disclosure: [email protected].

8. Responsible AI and employment law

Obi does not make people decisions. It doesn’t hire, fire, or promote — it prepares the human who does. Obi is trained not to make decisions even when asked, and PI monitors that guardrail. Using Obi is always optional.

Automated employment decision tools: PI’s position is that its products are not AEDTs under current definitions, and its tools should be one factor among many in any employment decision. Definitions vary by jurisdiction — review with your own counsel.

EU AI Act: PI is preparing for requirements under the Act with governance aligned to ISO 42001.

Accuracy: Obi draws only on PI-validated science; answers are checked against a formal PI rubric. Like any AI, it can make mistakes — review answers before acting on them.

9. Controls you have

Turn Obi off entirely — Your admin can do so at any time via AI Settings in PI Administration, without affecting other PI products.

Control company context — Only admins can add or change the optional context (website link, strategy text, up to 3 files).

Control who sees what — Role-based access and data segregation apply to Obi exactly as they do to the rest of PI.

After turning Obi off — Disablement applies going forward. It does not retroactively delete previously collected data; standard retention and deletion terms apply.

10. The agreements that apply

Document	What it covers
Client Software Agreement (predictiveindex.com/terms)	All PI services. Your prompts are Client Data — you own them, and grant PI the rights needed to provide the products and services.
AI Terms Addendum (predictiveindex.com/ai-terms)	AI-specific terms: rights you grant PI and its AI suppliers, AI accuracy, how prompts may persist within trained models, and your responsibilities (e.g., no regulated personal data beyond intended use; human review of consequential decisions).
Services Privacy Policy	All data you share with PI, including through AI products — what personal data is collected and how it’s used.
Data Processing Agreement	All personal-data processing through the platform, including AI features. Includes SCCs for EU data and defines breach-notification timelines (entered into with clients where applicable — typically EU, UK, and California). Via your PI account team.