Who decides on AI

The organization

Organization: Redfern State Department of Professional Licensing (fictional)
Industry: State government: issues and disciplines professional licenses
Size: About 450 staff; State IT runs the department's systems; a three-person security office
Risk appetite: Low tolerance for exposure of nonpublic applicant personal data; moderate tolerance for slower administrative work.

Constraints

The Deputy Director of Licensing owns applicant records and decides how they may be used.
State IT is the custodian: it runs the department's systems, web filtering and the state-contracted enterprise AI workspace.
The state already contracts for an enterprise AI workspace that doesn't train on inputs and has logging and retention controls. The department hasn't turned it on.
Under department policy the security office advises on risk but can't accept it; the data owner or the Agency Director does.
The security office has proposed blocking all AI tools and says it will sign off on the result.

Remediation capacity: 2 findings. You can choose "Mitigate: remediate ("fix")" for at most 2.

You have 2 remediation slots.

F1 Licensing staff paste applicant records into a free public AI chatbot

Web filter logs show about 300 sessions a week from licensing staff to a free public AI chatbot. Staff say they paste in case files, including applicant names, license numbers and background-check details, to get summaries. The free service's terms let it keep inputs and use them to improve its models. State IT can turn on the state-contracted enterprise AI workspace for licensing staff in about two weeks and then block the public site.

Asset: Applicant case files
Exposure: Third party
CVSS base score: None (not a scored vulnerability)
Known to be exploited: No
Data involved: Applicant personal data, including background-check results
Compensating controls available: Block the public chatbot at the web filter
Remind staff of the data-handling policy
Remediation effort: Medium
Business owner: Deputy Director of Licensing

Your response to F1

Avoid — stop the activity

Mitigate: remediate ("fix")

Mitigate: compensating control ("reduce")

Transfer / share

Who approves accepting this risk? (only if you choose Accept)

Your rationale (required if you accept: 10–600 characters; optional otherwise)

F2 AI notetaker extension records closed disciplinary hearings

Some hearing officers installed a free AI notetaker browser extension. It records closed disciplinary hearings held over video and stores the audio and transcripts in its vendor's cloud. The department has no contract with that vendor. A stenographer already produces the official record of every hearing. The hearing officers say the extension's summaries save them about 20 minutes per hearing.

Asset: Closed disciplinary hearings
Exposure: Third party
CVSS base score: None (not a scored vulnerability)
Known to be exploited: No
Data involved: Testimony and personal details from closed disciplinary hearings
Compensating controls available: Limit which browser extensions staff can install
Remediation effort: Low
Business owner: Deputy Director of Licensing

Your response to F2

Avoid — stop the activity

Mitigate: remediate ("fix")

Mitigate: compensating control ("reduce")

Transfer / share

Who approves accepting this risk? (only if you choose Accept)

Your rationale (required if you accept: 10–600 characters; optional otherwise)

F3 Communications drafts press releases with a public AI tool

The communications team uses a free public AI tool to draft press releases and social media posts. It works only from information the department has already published, such as board meeting results and public disciplinary orders. No written rule says what staff may enter. The Communications Director wants the team to keep using it.

Asset: Press release drafting
Exposure: Business process
CVSS base score: None (not a scored vulnerability)
Known to be exploited: No
Data involved: Information the department has already made public
Compensating controls available: Written rule: no nonpublic data in public AI tools
A person reviews every draft before release
Remediation effort: Low
Business owner: Communications Director

Your response to F3

Avoid — stop the activity

Mitigate: remediate ("fix")

Mitigate: compensating control ("reduce")

Transfer / share

Who approves accepting this risk? (only if you choose Accept)

Your rationale (required if you accept: 10–600 characters; optional otherwise)

F4 Case-management vendor turned on an AI assistant by default

The department's case-management software-as-a-service vendor released an AI assistant and turned it on for all customers. It sends case notes to a sub-processor to generate summaries. The contract predates the feature and says nothing about AI, training on customer data, sub-processors or breach notification for them. The vendor has offered to negotiate an amendment, and an admin setting turns the assistant off.

Asset: Case-management system
Exposure: Third party
CVSS base score: None (not a scored vulnerability)
Known to be exploited: No
Data involved: Case notes on applicants and licensees
Compensating controls available: Turn the AI assistant off until the contract is amended
Remediation effort: Medium
Business owner: Deputy Director of Licensing

Your response to F4

Avoid — stop the activity

Mitigate: remediate ("fix")

Mitigate: compensating control ("reduce")

Transfer / share

Who approves accepting this risk? (only if you choose Accept)

Your rationale (required if you accept: 10–600 characters; optional otherwise)

F5 Supervisor's spreadsheet uses AI to flag applicants for denial

A licensing supervisor built a spreadsheet that sends applicant data to a commercial AI service using a key on their personal account. The AI flags applications it rates as likely to be denied, and staff have started treating the flags as decisions. There's no documented human review. Nobody at the department knows the service's retention terms. Once the enterprise workspace from F1 is running, moving the spreadsheet into it takes about a week.

Asset: Applicant screening spreadsheet
Exposure: Business process
CVSS base score: None (not a scored vulnerability)
Known to be exploited: No
Data involved: Applicant personal data and application history
Compensating controls available: Require a documented human review of every flag before any action
Remediation effort: Medium
Business owner: Deputy Director of Licensing

Your response to F5

Avoid — stop the activity

Mitigate: remediate ("fix")

Mitigate: compensating control ("reduce")

Transfer / share

Who approves accepting this risk? (only if you choose Accept)

Your rationale (required if you accept: 10–600 characters; optional otherwise)

Overall note (optional, up to 1200 characters; shown back to you, never stored)

Your rationale and note are shown back to you once and never stored. Privacy