3rdComply
    User Manual

    AI User Manual & FAQ

    This manual explains how the AI engine within 3rdComply operates, what it is designed to do, how to use each AI-powered feature, how human oversight works, and how to override or report AI outputs. All AI outputs are advisory — human review is always required.

    AI Features

    3

    Sections

    10

    01

    Section 1

    What the AI Does (and Does Not Do)

    Intended Purpose

    The AI engine in 3rdComply focuses on benchmarking external vendor security postures against internal organizational baselines. It functions to save research time and provide context-aware drafting aids for vendor risk management.

    Core Boundaries

    Every AI output within 3rdComply is classified strictly as an advisory recommendation. The AI never takes autonomous action on your behalf, nor does it replace professional human expertise and legal counsel.

    The AI IS / DOES...

    The AI IS / DOES...Details
    A benchmarking aidIt extracts expected standards from internal policies for direct comparison.
    An automated researcherIt scrapes public vendor security portals using targeted search grounding.
    A risk analyzerIt evaluates vendor questionnaires against compliance criteria.

    The AI IS NOT / DOES NOT...

    The AI IS NOT / DOES NOT...Details
    A vendor disqualifierIt does not automatically block or blacklist vendors based on scores.
    A policy creatorIt does not draft new organizational rules, only extracts existing ones.
    A substitute for human due diligenceIt cannot verify if a vendor is lying; all outputs require human verification.

    ⚠ Mandatory Rule

    Never use AI output as the sole basis for a compliance decision. AI-generated findings, risk scores, and assessments must be reviewed, validated, and approved by a qualified human before any organizational action is taken.

    02

    Section 2

    AI Features per Product

    The 3rdComply platform leverages three primary AI capabilities designed to streamline vendor security alignment and cross-referencing:

    Web Search Query

    Queries restricted vendor domains using search grounding to complete unanswered security questions.

    Vendor Standard Benchmark

    Extracts expected security standards from internal policies to baseline questions.

    Questionnaire Evaluator

    Assesses external vendor answers against best practices to flag weak areas and suggest risk tiers.

    03

    Section 3

    Operational Feature Guides

    Web Search Query

    Purpose

    Query the web (restricted to the vendor's official domain/subdomains) to answer unanswered vendor questionnaire questions.

    Input

    Vendor name, URL, and questions.

    Output

    Grounded answers with cited URLs.

    How to Use

    1. 1Navigate to Vendor Portals and open a vendor assessment questionnaire.
    2. 2Select any unanswered questions and click Run Web Search.
    3. 3Specify the vendor's name and official domain URL (e.g., https://vendor.com).
    4. 4The AI executes a targeted search using search grounding, answers the questions based on the vendor's public security pages, and attaches direct citation links (URLs).

    Vendor Standard Benchmark

    Purpose

    Extract company security standards from internal policies to benchmark vendor questionnaire questions.

    Input

    Internal policy files + questions.

    Output

    Expected standards list.

    How to Use

    1. 1Navigate to Standards Setup.
    2. 2Upload company internal policies (PDF, DOCX, or XLSX).
    3. 3Input or select the security questions you wish to evaluate.
    4. 4Click Extract Standards Benchmark.
    5. 5The AI scans the policies, extracts the company's expected security baselines, and returns the expected standards for each question.

    Questionnaire Evaluator

    Purpose

    Assess third-party vendor questionnaire responses and suggest risk ratings.

    Input

    Completed vendor questionnaire.

    Output

    Risk scores + rationales.

    How to Use

    1. 1Navigate to Vendor Portals and open a completed vendor questionnaire.
    2. 2Click Analyze Answers.
    3. 3The AI reads each vendor response, evaluates it against compliance best practices, flags weak areas, and calculates a suggested risk score (Low, Medium, High).
    04

    Section 4

    Human Oversight & Mandatory Checkpoints

    All AI outputs are advisory recommendations and require explicit human intervention before being finalized or acted upon.

    Checkpoint Matrix

    The following checkpoints must be cleared by qualified personnel before moving forward in the workflow:

    FeatureCheckpointWhat to ReviewMinimum Reviewer Qualification
    Web Search QueryBefore locking in answers.Verify the cited URLs actually exist and support the answer.Any compliance team member.
    Vendor Standard BenchmarkBefore finalizing baseline.Verify that extracted standards match actual internal policies.Compliance Officer or auditor with framework expertise.
    Questionnaire EvaluatorBefore final vendor rating.Validate ratings; manually review any vendor scoring below 70%.Compliance Officer / Risk Analyst.

    Universal Evaluation Checklist

    When executing any checkpoint review, always assess the output against these parameters:

    Factual Accuracy

    Are the extracted rules, text quotes, and vendor references correct?

    Organizational Context

    Does the output account for your specific organization's size, industry, and risk profile?

    Recency

    Is the information gathered up to date, or does it reference outdated domain policies?

    Completeness

    Does the output address all aspects of the evaluation prompt, or are there gaps?

    05

    Section 5

    System Override Procedures

    The 3rdComply platform operates on the Override Principle: you always have the final say. The AI never takes autonomous action.

    Overriding Web Search Query Answers

    Modify Content

    Adjust the reasoning, modify the returned answer text, or correct/replace the cited URLs manually in the questionnaire editing pane.

    Overriding Vendor Standard Benchmark Settings

    Edit Extracted Baselines

    In the standards editor table, manually edit the extracted expected standard text, or delete a recommended standard and enter a custom requirement.

    Overriding Questionnaire Evaluator Scores

    Manual Risk Override

    Select a different risk level from the risk dropdown in the vendor profile interface. Enter a mandatory text justification for the manual override.

    Edit Rationale

    Click the edit icon on the assessment field to rewrite or clear the AI-suggested risk rationale.

    06

    Section 6

    Reporting AI Concerns

    All users, customers, partners, contractors, and other stakeholders are encouraged to promptly report any AI-related concerns, including but not limited to:

    • Unexpected, unintended, or harmful AI behavior.
    • Biased, discriminatory, or unfair outputs.
    • Inaccurate, misleading, deceptive, or unsafe responses.
    • Suspected privacy, security, or confidentiality incidents, including exposure of personal data or sensitive information.
    • Unauthorized access or misuse of AI systems.
    • Suspected non-compliance with applicable laws, regulations, contractual obligations, or organizational AI governance requirements.
    07

    Section 7

    Reporting Channels

    AI-related concerns may be reported through the appropriate communication channels or by contacting our Data Protection Officer (DPO) at privacy@feha.io. Reports should include, where possible, the affected AI system, a description of the issue, and any supporting evidence.

    All reported concerns are logged, assigned a unique reference, and tracked throughout investigation, remediation, and closure. Anonymous reporting may be accepted where applicable.

    08

    Section 8

    What to Report?

    When you encounter an AI output that is incorrect, unhelpful, or concerning, navigate to Feedback to submit a star rating (1–5) and a text description. Use the following guide to format what to include:

    SituationWhat to Include in Feedback
    Factual ErrorThe incorrect claim, the correct answer, and the source.
    HallucinationThe fabricated control/article ID or website quote and confirmation it doesn't exist.
    BiasThe query, the framework/baseline you expected, and what you got.
    Security ConcernIf the AI revealed information it shouldn't have, or bypassed guardrails.
    09

    Section 9

    FAQ (Frequently Asked Questions)

    Q1: Is my data sent to external AI providers?

    Yes, document content is sent to a third-party API for processing. However, all PII (names, emails, phone numbers, credit cards, API keys, etc.) is automatically masked before sending. It does not use API data for training; your original data is processed transiently and discarded after the request completes.

    Q2: Can the AI access my company's internal data?

    Only if you explicitly upload it. The AI has access strictly to: (1) the curated regulatory knowledge base (public frameworks), and (2) documents you upload during the current session. It cannot access your MySQL database, your other FEHA data, or other companies' data.

    Q3: What happens if I accidentally upload a file with sensitive data?

    The PII masking system (Presidio + 11 custom recognizers) automatically detects and masks 22 types of sensitive data before sending it to the LLM. Your original file is processed transiently—it is never stored in the AI system. After processing completes, all uploaded content is discarded.

    Q4: Can someone prompt-inject the AI through my documents?

    The system has guardrails that detect 40+ categories of prompt injection patterns in both user input and uploaded document content. Malicious content is rejected with a generic error message. The system also sanitizes retrieved RAG context to prevent injection from the knowledge base.

    Q5: What file formats can I upload?

    PDF, DOCX, XLSX, XLS, JSON, PNG, JPG, TIFF. Each file is validated for both file extension and magic byte signature (the system checks the actual file contents, not just the name). The file upload limit is a maximum of 10 MB per file.

    Q6: I got an error saying "inappropriate content detected." What happened?

    Your input triggered the guardrail system. This can happen if your text contains patterns that resemble prompt injection, even unintentionally. Try rephrasing your question. If you believe this is a false positive, report it via the Feedback feature.

    Q7: Are my AI interactions logged?

    Yes. Each AI request logs: model used, message count, input preview, PII entities detected, response preview, and token usage. These logs are used for quality monitoring and incident investigation. They are not shared externally. Observability is provided through Langfuse.

    Q8: The evaluation generation is taking a very long time to respond.

    For file analysis features or background operations like compilation modules, the expected processing time is 30–120 seconds depending on file count and size. Check the job status indicator. If persistent, contact your system administrator to check the AI Services container health.

    10

    Section 10

    Key Reporting & Support Contact

    In accordance with your privacy and data processing frameworks, any suspected model bias, security concerns, or data processing violations can be directed to the Privacy Contact / Data Protection Representatives:

    Privacy Contact / Data Protection Representative (DPO)

    Henry Kevin Marcelino Ratu

    Ryan Yosedie

    privacy@feha.io

    All AI outputs within 3rdComply are classified as advisory recommendations. Never use AI output as the sole basis for a compliance decision. For bias reports, security concerns, or AI-related enquiries, contact us at privacy@feha.io.