Skip to main content

    Sample report — Juniper Claims Services

    QuickScan sample report

    Company names and figures are invented.

    These observations come from your answers, using fixed rules in this browser. No controls have been verified. Each area is shown separately, with critical concerns listed on their own and no overall score. This is preliminary guidance, not an audit, certification or legal conclusion.

    Critical concerns

    • Shared or uncontrolled AI credentials

      Shared accounts and unrotated API keys make AI activity unattributable and are a direct route for misuse after people leave or keys leak.

      Raised by your answer: “Shared accounts or uncontrolled keys are in use”. Critical concerns are never offset by other answers.

    Observations by area

    Each area's level is its weakest applicable answer. Open “Why you see this” on any card to see the answer that set it.

    AI system inventory

    Partial

    Practices exist in part or informally. The gap is usually documentation and consistency rather than absence; the items below show what would close it.

    Why you see this
    • Do you have a list of the AI systems, tools and agents in use — including AI features inside existing vendor products and tools staff adopted on their own?

      Your answer: A partial list exists, or it is not kept up to datesets the level

    • For each listed AI system, is the business purpose, the data it uses and a named owner recorded?

      Your answer: For some entriessets the level

    • Rule: the weakest applicable answer sets the area level (Concern › Unknown › Partial › Evidence available). Not applicable answers are left out.

    Ownership and accountability

    Partial

    Practices exist in part or informally. The gap is usually documentation and consistency rather than absence; the items below show what would close it.

    Why you see this
    • Is a named person accountable for how AI is used in the organization?

      Your answer: Someone looks after it informally; it is not written downsets the level

    • Is there an approved AI use policy (or AI section in an existing policy) that staff can find?

      Your answer: A draft exists, or it was approved but not communicatedsets the level

    • Rule: the weakest applicable answer sets the area level (Concern › Unknown › Partial › Evidence available). Not applicable answers are left out.

    Authorization of AI tasks and agents

    Partial

    Practices exist in part or informally. The gap is usually documentation and consistency rather than absence; the items below show what would close it.

    Why you see this
    • Before a new AI use case, tool or agent goes live, is there a documented approval step?

      Your answer: Sometimes, or only for some teamssets the level

    • Do AI agents or automations take actions in your systems (send, change, approve, pay, delete)? If so, are the permitted actions defined and limited?

      Your answer: Yes they act — limits are partly defined or rely on the vendor's defaultssets the level

    • Rule: the weakest applicable answer sets the area level (Concern › Unknown › Partial › Evidence available). Not applicable answers are left out.

    Data handling

    Partial

    Practices exist in part or informally. The gap is usually documentation and consistency rather than absence; the items below show what would close it.

    Why you see this
    • Are there rules on which data may be entered into AI tools (customer, personal, confidential or regulated data)?

      Your answer: Rules exist but depend on staff judgmentsets the level

    • Where AI output feeds a decision or customer-facing content, is it reviewed for accuracy and is the review recorded?

      Your answer: Reviewed sometimes, or without a recordsets the level

    • Rule: the weakest applicable answer sets the area level (Concern › Unknown › Partial › Evidence available). Not applicable answers are left out.

    Vendors and access

    Concern

    At least one practice in this area is missing according to your answers. This is the weakest answer in the area and sets its level; the other answers do not offset it.

    Why you see this
    • Are AI vendors — and AI features switched on inside existing vendors — reviewed before use (terms, training on your data, sub-processors, retention)?

      Your answer: Reviewed informally, or only for major vendors

    • Is access to AI tools, API keys and agent credentials controlled — role-based, removed when people leave, keys rotated?

      Your answer: Shared accounts or uncontrolled keys are in usesets the level

    • Rule: the weakest applicable answer sets the area level (Concern › Unknown › Partial › Evidence available). Not applicable answers are left out.

    Human oversight

    Partial

    Practices exist in part or informally. The gap is usually documentation and consistency rather than absence; the items below show what would close it.

    Why you see this
    • For AI-supported decisions that affect people, money or safety, does a person review the outcome before it takes effect?

      Your answer: Yes — defined for those decisions and the review is recorded

    • Can you pause or switch off an AI system or agent quickly, and has that been tested?

      Your answer: Probably — it has not been testedsets the level

    • Rule: the weakest applicable answer sets the area level (Concern › Unknown › Partial › Evidence available). Not applicable answers are left out.

    Incidents

    Partial

    Practices exist in part or informally. The gap is usually documentation and consistency rather than absence; the items below show what would close it.

    Why you see this
    • Do staff know how to report an AI-related problem (wrong output acted on, data entered by mistake, an agent doing something unexpected), and are reports logged?

      Your answer: People would tell someone; nothing is defined or loggedsets the level

    • In the last 12 months, has any AI-related incident or near miss occurred, and what happened to it?

      Your answer: None that we know of, and we have a way to find outsets the level

    • Rule: the weakest applicable answer sets the area level (Concern › Unknown › Partial › Evidence available). Not applicable answers are left out.

    Evidence availability

    Unknown

    You answered Unknown to a question in this area. Unknown is not counted as a pass; finding out is the first step.

    Why you see this
    • If an assessor asked today for the inventory, policy, approvals, vendor reviews and oversight records, how much could you produce within a few days?

      Your answer: Some of it

    • Are AI system activity logs (prompts, actions, outputs where appropriate) retained with a defined retention period and accessible to you?

      Your answer: Unknown — I don't know or could not find outsets the level

    • Rule: the weakest applicable answer sets the area level (Concern › Unknown › Partial › Evidence available). Not applicable answers are left out.

    Three next actions

    Chosen by rule, in this order: critical concerns, areas of concern, unknown areas, partial areas, then general upkeep. Always three — not a ranking of everything you could do.

    1. Priority 1

      Replace shared AI accounts and rotate uncontrolled keys

      Give each person their own account, assign an owner to each API key and agent credential, rotate the ones you cannot attribute, and add AI tools to the checklist used when employees leave the organization.

      BecauseYou answered "Shared accounts or uncontrolled keys are in use" — flagged as a critical concern.

    2. Priority 2

      Find out the answers you marked Unknown (evidence availability)

      Ask the people who run the systems. Unknown answers are the easiest gaps to close, and they determine whether the other observations hold.

      BecauseYou answered Unknown or left questions blank in one area.

    3. Priority 3

      Build or complete the AI system inventory

      Ask every team which AI tools, vendor AI features and agents they use and record each one. The free AI System Inventory Template on the Resources page has the columns. Set a review date so the list stays current.

      BecauseAI system inventory: you answered "A partial list exists, or it is not kept up to date".

    Evidence an assessor would ask for

    Listed for every applicable area, whatever its level — “Evidence available” means you said something could be shown, not that it was seen. Use this as a gathering list; the free Audit Evidence Checklist goes deeper.

    AI system inventory

    Partial
    • Current AI system inventory (including vendor AI features and employee-adopted tools)
    • For each entry: purpose, data used, owner, go-live date and status
    • Date and record of the last inventory review

    Ownership and accountability

    Partial
    • Name and role of the person accountable for AI use, and where that is recorded
    • Approved AI use policy with approval date and distribution record
    • Terms of reference or minutes of the group that oversees AI use, if one exists

    Authorization of AI tasks and agents

    Partial
    • Approval records for AI use cases, tools and agents before go-live
    • For each agent or automation: the permitted actions, systems reached and limits
    • Change records showing how permissions or scope were extended

    Data handling

    Partial
    • Rules on which data classes may be entered into which AI tools
    • Technical controls that enforce those rules (DLP, allow-lists, enterprise tenants)
    • Records of output review for AI-supported decisions or published content

    Vendors and access

    Concern
    • Vendor review records: terms, use of your data for training, sub-processors, retention
    • Access list for AI tools, API keys and agent credentials with owners
    • Onboarding, departure and key-rotation records for AI credentials

    Human oversight

    Partial
    • Definition of which AI-supported decisions require human review before taking effect
    • Records of that review (who, when, outcome, overrides)
    • Documented and tested procedure to pause or switch off an AI system or agent

    Incidents

    Partial
    • Reporting route for AI-related problems and the record that staff were told about it
    • AI incident and near-miss log with follow-up actions
    • Link between the AI reporting route and the existing security/privacy incident process

    Evidence availability

    Unknown
    • Where the documents above are kept and who can retrieve them
    • Log retention settings for AI systems (prompts, actions, outputs where appropriate)
    • A sample log extract that shows the records are accessible

    How this result was produced

    Fixed rules, applied in your browser: the weakest applicable answer sets each area's level; unanswered questions count as Unknown; Unknown and Not applicable are never a pass; critical concerns come from specific answers and are listed separately; the three actions are chosen in a fixed order. No percentage or overall score is calculated and no AI model is used.

    These fictional answers for Juniper Claims Services show how the rules produce observations and actions. Questions follow the governance themes of the NIST AI Risk Management Framework. For a professional evidence review, explore AI & Agentic AI Assurance.

    This is the fictional example. Run the QuickScan on your own organization — it stays in your browser.

    Start the QuickScan