This Policy describes how Pansophy AI Inc. designs, deploys, and governs the artificial intelligence capabilities in the Second Opinion. It is provided in accordance with Microsoft's Responsible AI principles and applicable regulatory requirements governing AI systems used in workplace contexts.
This Responsible AI and Transparency Policy applies to all users and organizations that access the Second Opinion through Microsoft Teams and the Microsoft Marketplace. It governs how AI capabilities are designed, what they can and cannot do, and the obligations of both Pansophy AI and its users in ensuring responsible use.
Second Opinion uses artificial intelligence to analyze language patterns in documents you submit. By using the Software, you acknowledge and understand the following:
Second Opinion does not make recommendations of any kind. All outputs require human review. The human reviewer retains full decision authority at all times.
The Software is intended to:
Second Opinion must not be used as the sole or primary basis for any consequential decision about a person; used to automate decisions without human review; or used to evaluate individuals on the basis of any protected characteristic under applicable law.
The most significant bias mitigation in Pansophy AI is architectural. The analysis engine processes only anonymized word patterns. It receives no demographic inputs of any kind. It cannot see who is speaking. It has no knowledge of the speaker's age, gender, race, ethnicity, national origin, disability status, or any other personal characteristic. This is not a limitation of the current version; it is a deliberate design principle that will not change.
This architecture is fundamentally different from machine learning-based assessment tools that train on historical outcomes. Second Opinion has no feedback loop from past decisions and no mechanism for demographic inference. The model cannot learn to discriminate because it never sees protected class data.
The analysis methodology is derived from the Linguistic Inquiry and Word Count (LIWC) research tradition, a word-count-based psycholinguistic approach that has been in academic and commercial use for decades. It counts word categories in text. It does not know who wrote or spoke those words.
Pansophy AI acknowledges the following limitations inherent in any linguistic analysis approach:
In addition to the lexicon-blind architecture, Pansophy AI implements the following measures to reduce bias risk:
Pansophy AI commits to ongoing monitoring of its AI systems for disparate impact and bias. We review aggregated, anonymized output patterns periodically and update our interpretation frameworks where concerns are identified.
All data processed by the Software is stored and processed exclusively within Microsoft Azure's United States data centers (East US 2 region). No data is transmitted to or stored in non-Microsoft cloud environments. Second Opinion is available only to Microsoft 365 tenants based in the United States.
Personally identifiable information is identified and redacted from extracted text prior to behavioral scoring, using Microsoft Azure AI Language. This step runs automatically on every document submission. The PII-redacted text is what reaches the scoring engine and the language model.
Pansophy AI applies a split retention model. Uploaded files, extracted text, and PII-redacted text are discarded at session close and are not written to persistent storage. Derived trait scores and the written narrative are retained by Pansophy AI for legal defensibility, model refinement, and ordinary business operations. Full retention detail is in the Privacy Policy.
Users have the right of access, deletion, correction, and portability with respect to their data. To exercise these rights, contact privacy@pansophyai.com. Requests will be actioned within seven (7) business days.
You are solely responsible for obtaining required consent from individuals whose words are contained in documents you submit. Pansophy AI does not obtain consent from subjects directly.
The Software is designed to support compliance with applicable AI and data privacy regulatory frameworks. Where the Software is used in connection with employment decisions, users should additionally be aware of and ensure compliance with the following as applicable to their jurisdiction:
Second Opinion is available only to Microsoft 365 tenants based in the United States; cross-border processing under GDPR is therefore not applicable to this Software.
Pansophy AI does not provide legal advice. Organizations should consult qualified legal counsel to assess compliance obligations specific to their jurisdiction and use case.
Pansophy AI is built on the Microsoft Azure platform and is designed in alignment with Microsoft's six Responsible AI principles:
Descriptive outputs only. No individual is ranked or rated relative to others. Recommendation language is programmatically excluded.
Operates exclusively within the Microsoft Azure sovereignty boundary. Outputs include contextual framing to support accurate interpretation.
All data remains within the user's Microsoft tenant. PII scrubbing applied prior to external processing. AES-256 encryption at rest.
Onboarding materials disclose limitations for non-native English speakers and diverse communication styles.
This Policy and the onboarding flow disclose the nature, capabilities, and limitations of the AI system to all users before first use.
Human oversight required for all consequential decisions. Ongoing monitoring of outputs for bias and accuracy. Annual policy review.
The following disclosure is presented to all first-time users during the Software onboarding flow and is prepended to every written narrative generated by the Software:
This opinion is based solely on word patterns in the provided text using a validated language scoring model. The system has no access to and does not process any demographic or PII information about the subject. This opinion is intended as one input among multiple documented factors in any decision made, including an employment decision. Second Opinion does not make recommendations. We challenge your assumptions and confirm your hunches.
This disclosure is also available on demand within the agent at any time by asking "How does this analysis work?"
This Policy is reviewed and updated at least annually and whenever material changes are made to the Software's AI capabilities. The current version is available at this URL.
Questions regarding this Policy or requests to exercise data subject rights should be directed to: