Artificial Intelligence Usage Policy

Purpose

Core 12 uses artificial intelligence tools to improve agency delivery, software development, documentation, analysis, automation, and internal operations. This policy permits useful AI adoption while protecting client information, Core 12 information, source code, credentials, intellectual property, and compliance evidence.

Policy

AI tools may be used when the user remains accountable for the work, data is handled according to classification, and outputs are reviewed before they affect a client, production system, security decision, contract, financial record, or personnel decision.

AI output is not authoritative by default. Employees and contractors are responsible for validating facts, security implications, code behavior, licensing, client requirements, and brand or legal claims before use.

AI-related work must stay within the user's current role, position, assigned client work, and authorized Core 12 responsibilities. Using AI to work on concepts, opportunities, automations, repositories, systems, or projects outside the user's role requires approval from senior management: the IT Director, CEO, COO, or CFO.

Scope

This policy applies to employees, contractors, interns, and approved third parties using AI tools for Core 12 or client work.

AI tools include chat assistants, coding agents, code-completion tools, image or media generators, transcription and meeting-summary tools, search or research assistants, workflow automations, AI APIs, and agentic tools that can read files, write files, access systems, call APIs, or take actions.

Reference Frameworks

This policy is informed by the NIST AI Risk Management Framework, the NIST Generative AI Profile, OWASP Top 10 for Large Language Model Applications, and ISO/IEC 42001. Core 12 is not claiming ISO/IEC 42001 certification unless separately approved and certified.

Allowed Use

AI is approved for routine productivity work when the data involved is Public or properly controlled Internal information:

• Brainstorming, outlining, rewriting, summarizing, and formatting content.
• Drafting internal documentation, policies, templates, and checklists.
• Creating first drafts of client-facing content that will receive human review.
• Reviewing code, explaining code, proposing tests, writing non-production scripts, and helping with implementation.
• Summarizing non-sensitive meetings, notes, and research.
• Creating task lists, issue drafts, migration plans, and operating procedures.
• Producing design, copy, and technical recommendations that will be reviewed by an accountable person.

Controlled Use

The following uses are allowed only with additional controls:

Use	Required controls
Client deliverables	Human review for accuracy, brand fit, contractual fit, and confidentiality before delivery.
Client source code or private repositories	Approved tool, least-privilege repository access, no secrets in prompts, code review, tests, and security checks before merge or deployment.
Client documents or business information	Data minimization, classification check, approved tool terms, and client authorization or contractual basis where required.
Internal Confidential information	Approved business account, training/data-retention controls where available, and a documented business need.
AI agents with file, repository, browser, shell, or API access	Work in controlled workspaces, use least privilege, keep logs/devlogs/issues, and require human review before production-impacting changes.
Recurring AI automations	Register the tool/service and workflow in core12-isms-management before operational use.
AI-enabled applications, processes, scripts, integrations, or unattended workflows used in a company process	Use a company-owned AI platform account or company-owned API key, with owner, access, logging, billing, and offboarding controls recorded before operational use.
AI-generated media for clients	Review licensing, usage rights, brand suitability, disclosure requirements, and client approval requirements before publication.
Personal or reimbursed AI subscriptions connected to Core 12 services	Register the tool, use Core 12-controlled service access, prohibit copying work to personal spaces, and include the account in access review and offboarding.

Personal or Reimbursed AI Subscriptions

Core 12 may allow employees or contractors to use personal AI subscriptions for Core 12 work, including subscriptions reimbursed by the company, when the AI service can be used without weakening Core 12 control over company or client information.

Reimbursement does not make a personal AI account a company-controlled system. When a personal or reimbursed AI account connects to Core 12 services, the connected access must remain controlled by Core 12.

Personal or reimbursed AI subscriptions may connect to Core 12 services only when all of the following are true:

1. The AI tool and account use case are recorded in core12-isms-management.
2. The connected Core 12 service access uses the user's authorized Core 12 identity, not a personal email identity.
3. OAuth grants, API tokens, repository access, file access, browser access, or other connectors are approved, least-privilege, and revocable by Core 12.
4. The AI tool does not duplicate, synchronize, index, train on, retain, or move Core 12 or client information into a personal workspace beyond the approved task.
5. Official work products are stored in Core 12-approved systems, not left as the only copy in the AI account or personal storage.
6. Confidential information, client information, personal data, source code, or regulated information is used only when the tool configuration and vendor terms are approved for that classification.
7. The account is included in access reviews when it has ongoing access to Core 12 services.
8. Offboarding includes revoking AI tool tokens, OAuth grants, browser sessions, repository permissions, shared links, and any other connected access.

If these conditions cannot be met, the AI tool must be used only with sanitized Public or Internal information that does not identify clients, expose Core 12 assets, or create a retained company work product in a personal space.

Company AI Platform Requirement

Applications, processes, scripts, integrations, agents, automations, or other work that use AI or automation as part of a Core 12 company process must use a company-owned AI platform account or company-owned API key.

This requirement also applies when the AI-enabled work acts without explicit human review before each action or output. Examples include background scripts, scheduled jobs, client or internal workflow automations, autonomous agents, production or publishing pipelines, data-processing jobs, decision-support workflows, and integrations that read from or write to company or client systems.

Personal, personal-reimbursed, employee-owned, or vendor-trial AI accounts must not be used as the execution account or API key for company-process automation, unattended AI workflows, or AI-enabled work that acts before review. They may be used only for reviewed individual productivity work when the requirements in this policy are met.

Company AI platform accounts and API keys must:

1. Be approved for the intended use case, data classification, and systems involved.
2. Be recorded in core12-isms-management with business owner, technical owner, use case, permitted data, connected systems, review date, and offboarding steps.
3. Use least-privilege access and approved secret-management practices.
4. Support Core 12 control over billing, logs, usage visibility, revocation, retention settings, and vendor terms.
5. Be reviewed before the workflow is expanded to new clients, departments, systems, data classes, or unattended actions.

Role Scope and Senior Management Approval

AI tools must be used only for work that fits the user's current role, position, assigned accounts, approved projects, and authorized responsibilities.

Approval from the IT Director, CEO, COO, or CFO is required before using AI tools to:

1. Research, develop, prototype, or automate a business concept outside the user's role.
2. Access repositories, systems, client files, or business records outside the user's assigned work.
3. Create or operate AI agents that act across departments, clients, financial records, personnel records, legal/compliance work, or executive business planning.
4. Use Core 12 or client information to support a side project, personal project, vendor evaluation, new service concept, or unassigned sales opportunity.
5. Connect personal or reimbursed AI tools to Core 12 services for a use case that has not been approved.

Approval must be documented in core12-isms-management and include approver, scope, tool, data classification, systems involved, expiration or review date, and any compensating controls.

Prohibited Use

Users must not:

1. Submit passwords, API keys, private keys, session tokens, recovery codes, MFA seeds, database dumps, or production credentials to AI tools.
2. Submit Confidential client data, personal data, regulated data, or bulk production data to unapproved public or personal AI accounts.
3. Use AI to bypass access controls, monitoring, licensing, paywalls, security controls, or client restrictions.
4. Use AI to make final legal, financial, hiring, firing, promotion, disciplinary, credit, insurance, medical, or other consequential decisions without approved human decision-making and legal review.
5. Use AI-generated output as factual, legal, security, or compliance evidence without verification.
6. Allow an AI agent to deploy to production, delete data, change access rights, send external communications, or make purchases without explicit approval and review.
7. Create deceptive content, fake testimonials, false claims, impersonation, undisclosed synthetic evidence, or manipulative client/customer communications.
8. Use personal AI accounts for Core 12 or client Confidential information when an approved business tool is required.
9. Copy, move, synchronize, or retain Core 12 or client work products in personal AI workspaces, personal drives, personal notes, personal code repositories, or other personal systems.
10. Use AI tools to perform work outside the user's role, assigned projects, or authorized responsibilities without senior management approval.

Data Handling Requirements

Prompts, uploaded files, tool context, transcripts, retrieved documents, and AI outputs must be handled as information assets.

1. Classify the information before submitting it to an AI tool.
2. Use the minimum information needed for the task.
3. Remove or mask credentials, personal data, unnecessary client identifiers, and secrets before use.
4. Prefer approved business accounts with administrative controls, MFA/SSO where available, and training or retention controls suitable for the classification.
5. Store final work products in the approved source-of-truth location, not only in an AI chat history.
6. Record security-relevant AI decisions, code changes, review findings, or compliance actions in the appropriate repository, issue, devlog, or management record.
7. If information is accidentally submitted to an unapproved AI tool, report it as a security event the same business day.

Client Work Requirements

AI may support client work, but Core 12 remains accountable for the delivered result.

• Client-facing work must be reviewed by the responsible employee or contractor before delivery.
• Client requirements, contracts, confidentiality terms, brand rules, and regulatory requirements override this policy where stricter.
• Material AI use must be disclosed to the client when required by contract, law, platform rules, professional duty, or client expectation.
• AI-generated recommendations must be checked against the actual client environment and not presented as verified facts until verified.
• Client data must not be used to train, fine-tune, or improve an AI model unless the client has approved that use in writing and the Information Security Officer has approved the control set.

Software Development and Automation Requirements

AI-assisted development is allowed and expected for Core 12 agency work when normal engineering controls remain in place.

1. Source code generated or modified with AI must be reviewed by a human before merge.
2. Generated code must be tested according to the risk of the change.
3. Security-sensitive changes require security review appropriate to the system and data classification.
4. AI agents must not receive production secrets unless the agent, environment, and access path have been explicitly approved.
5. AI agents must operate in a controlled workspace or worktree when changing repository files.
6. Generated dependencies, licenses, and copied snippets must be reviewed for supply-chain and licensing risk.
7. AI-generated code must not be merged when the reviewer cannot explain what it does.
8. Automated AI workflows must include rollback, logging, ownership, and review controls appropriate to their impact.
9. AI-enabled applications, processes, scripts, integrations, agents, or automations that are part of a company process or act without explicit review must run through a company-owned AI platform account or company-owned API key.

Tool Approval and Inventory

Recurring or business-critical AI tools must be recorded in core12-isms-management before broad use.

At minimum, the record must identify:

• Tool name and vendor.
• Business owner.
• Approved use cases.
• Data classification allowed.
• Whether client data, personal data, source code, or Confidential information may be used.
• Account type and administrative controls.
• MFA/SSO availability.
• Data retention, training, and deletion controls.
• Contract, DPA, or vendor security review status where applicable.
• Whether the account is company-owned, personal, or reimbursed.
• Whether any company-owned AI platform account or API key is required because the tool supports a company process or acts without explicit review.
• Whether the tool connects to Core 12 services through OAuth, API tokens, browser sessions, repository access, files, email, calendars, chat, or cloud storage.
• Offboarding and access-revocation steps.
• Next review date.

Approved AI tools may be tracked as IT Service, Desktop Application, Licence, Vendor, Web Application, or Document Review records depending on the tool and use case.

Output Review Requirements

Before using AI output, the responsible person must review for:

• Accuracy and completeness.
• Confidentiality and accidental disclosure.
• Security vulnerabilities or unsafe instructions.
• Licensing, copyright, and attribution concerns.
• Client contract, brand, and accessibility requirements.
• Unsupported claims, fabricated citations, or fabricated evidence.
• Bias, unfairness, or inappropriate tone.

Training and Awareness

Personnel using AI tools must understand:

• This policy and the Data Classification Policy.
• Prompt and output handling expectations.
• Security risks such as prompt injection, sensitive information disclosure, insecure output handling, excessive agency, and supply-chain exposure.
• How to report accidental disclosure or suspicious AI behavior.

AI usage expectations must be included in security awareness training or periodic security reminders.

Exceptions

Exceptions require Information Security Officer approval and must be documented in core12-isms-management as a risk, control-gap issue, or approved exception record. Exceptions must include business reason, data involved, tool involved, compensating controls, owner, approval date, and review date.

Exceptions that involve work outside a user's role or position require approval from the IT Director, CEO, COO, or CFO.

Management Repo Use

• Register recurring AI tools and automations as IT Service, Desktop Application, Licence, Vendor, or Web Application records.
• Track company-owned AI platform accounts and API keys used by AI-enabled company processes, scripts, integrations, agents, or unattended workflows.
• Track personal or reimbursed AI subscriptions that connect to Core 12 services, including connector scope and offboarding controls.
• Track senior management approvals for AI work outside a user's role, position, assigned client work, or authorized responsibilities.
• Open risks for unapproved AI tools, unclear vendor terms, excessive data exposure, missing review controls, or agentic workflows with high impact.
• Open incidents for accidental disclosure of secrets, personal data, client Confidential information, or unauthorized AI tool use.
• Track policy reviews and exceptions through document-review and control-gap issues.
• Use migration tasks when legacy process material should become AI review checklists, training material, or register records.

Current Recommendation

Core 12 should treat AI as an approved productivity and delivery capability, not as shadow IT. The preferred operating model is:

1. Use approved business AI tools for agency work.
2. Allow personal or reimbursed AI subscriptions only when Core 12 controls connected service access and company/client work is not duplicated or moved into personal spaces.
3. Keep official work products in GitHub, approved documentation workspaces, client systems, or core12-isms-management.
4. Use AI aggressively for drafting, coding assistance, documentation, review preparation, migration planning, and automation.
5. Use company-owned AI platform accounts or company-owned API keys for AI-enabled company processes and any AI automation that acts without explicit human review.
6. Keep AI work aligned to the user's current role and require senior management approval for work outside that role.
7. Keep humans accountable for client commitments, security decisions, production changes, and published work.
8. Register recurring AI tools and automations so future review reminders, evidence collection, and audits can be automated.