1. Purpose

The purpose of this AI Acceptable Use Policy is to establish requirements for the responsible, secure, and appropriate use of artificial intelligence (AI) and generative AI tools within the organization.

The policy is intended to:

• Protect organizational, customer, employee, and confidential information.
• Reduce security, privacy, legal, and operational risks associated with AI use.
• Define which AI use cases are acceptable and which require approval.
• Establish requirements for using third-party AI services.
• Prevent unauthorized disclosure of sensitive information through AI tools.
• Ensure appropriate human review of AI-generated outputs.
• Support responsible adoption of AI while maintaining information security and compliance.

Core principle:

AI may be used to support business activities, but users remain responsible for protecting information, validating AI outputs, and complying with organizational requirements.

---

2. Scope

This policy applies to:

• Employees
• Contractors
• Consultants
• Interns
• Temporary personnel
• Third parties authorized to use organizational AI resources

It applies to AI systems used for:

• Text generation
• Code generation
• Data analysis
• Document summarization
• Translation
• Research
• Customer support
• Content creation
• Image generation
• Audio/video generation
• Automation
• Decision support
• Machine learning
• AI-powered SaaS applications
• AI APIs and models
• Internally developed AI systems

This policy applies whether AI is accessed through:

• Web applications
• Mobile applications
• Enterprise platforms
• APIs
• Cloud services
• Developer tools
• Integrated SaaS applications
• Internally hosted models

---

3. What Is AI?

For this policy, AI includes systems that can generate, analyze, classify, predict, recommend, summarize, or otherwise process information using machine-learning or artificial-intelligence techniques.

Examples include:

• Generative AI assistants
• Large language models
• AI coding assistants
• AI-powered productivity tools
• AI chatbots
• Machine-learning platforms
• AI analytics tools
• AI image/video generators
• AI-enabled SaaS applications

The organization may maintain an Approved AI Tools Register to identify AI services authorized for business use.

---

4. AI Use Principles

Users must follow these principles:

4.1 Authorized Use

Use only AI tools approved or permitted by the organization for the intended business purpose.

4.2 Data Protection

Do not provide information to an AI service unless the user is authorized to disclose that information to the service.

4.3 Minimum Necessary Information

Provide only the minimum information necessary to achieve the intended purpose.

4.4 Human Responsibility

AI output does not replace human accountability.

The employee using the AI system remains responsible for reviewing and appropriately using its output.

4.5 Accuracy

AI-generated information should be verified before being used for important business, security, legal, financial, customer, or technical decisions.

4.6 Security

AI tools must not be used to bypass organizational security controls.

4.7 Transparency

Where required, users should disclose the use of AI in accordance with organizational, customer, contractual, legal, or regulatory requirements.

---

5. Approved AI Tools

The organization should maintain a list of AI tools approved for business use.

Example:

AI Tool	Purpose	Approved Data	Restrictions	Owner	Status
Enterprise AI Assistant	Productivity	Internal information	No restricted data	IT	Approved
AI Coding Assistant	Development	Approved source code	No secrets/credentials	Engineering	Approved
Public AI Service	General research	Public information only	No confidential data	Security	Restricted
AI Customer Support Tool	Customer support	Approved customer data	Contract/security review required	Support	Approved

Approval should consider:

• Security
• Privacy
• Data handling
• Provider terms
• Data retention
• Model training/data-use practices
• Access control
• Encryption
• Contractual requirements
• Regulatory considerations
• Business purpose

---

6. Prohibited AI Use

Unless specifically authorized, users must not use AI tools to:

• Enter passwords or credentials.
• Enter API keys or access tokens.
• Enter encryption keys.
• Upload confidential customer information.
• Upload restricted organizational information.
• Upload sensitive employee information.
• Upload security incident information.
• Upload confidential contracts.
• Upload proprietary source code.
• Upload production databases.
• Upload security configurations that could expose organizational defenses.
• Circumvent security controls.
• Generate malware for unauthorized purposes.
• Conduct unauthorized attacks or penetration testing.
• Impersonate individuals.
• Create fraudulent content.
• Generate content for illegal activities.
• Make unauthorized decisions on behalf of the organization.

---

7. Data Classification and AI Use

AI usage must follow the organization’s information classification requirements.

A practical model is:

Data Classification	AI Usage
Public	Generally permitted for approved business use
Internal	Permitted only with approved AI tools
Confidential	Requires an approved tool and appropriate safeguards
Restricted	Prohibited unless specifically authorized

The organization should define its own classification and AI handling rules based on its risk assessment.

---

8. Customer Information

Customer information must not be entered into an AI service unless:

• The service is approved.
• The intended use is authorized.
• The data handling arrangements are understood.
• Applicable contractual requirements are satisfied.
• Applicable privacy/security requirements are addressed.

Example

An employee receives a confidential customer security report.

They should not copy the report into a public AI chatbot to obtain a summary.

If AI-assisted summarization is required:

Confirm authorization → Confirm approved AI service → Minimize data → Process → Review output → Secure/delete working data where required

---

9. Personal Data

Users must handle personal data in accordance with applicable privacy requirements and organizational policies.

Unless specifically authorized, users should not enter:

• Customer personal data
• Employee personal data
• Applicant information
• Health information
• Financial information
• Government identifiers
• Authentication information

into public or unapproved AI tools.

Where AI processing of personal data is approved, appropriate security, privacy, contractual, and retention requirements should be considered.

---

10. Source Code and AI Coding Tools

AI coding assistants may be used only in accordance with approved development practices.

Developers must:

• Use approved coding tools.
• Follow repository access controls.
• Avoid entering secrets into AI prompts.
• Avoid exposing customer data.
• Review AI-generated code.
• Conduct appropriate security testing.
• Check dependencies and licenses.
• Follow secure coding standards.
• Ensure generated code does not introduce security vulnerabilities.

AI-generated code must not automatically be treated as secure or production-ready.

Required Flow

Generate → Review → Security Check → Test → Approve → Deploy

---

11. Secrets and Credentials

The following must never be entered into an unapproved AI service:

• Passwords
• API keys
• Access tokens
• Private keys
• Encryption keys
• Database credentials
• AWS credentials
• Azure credentials
• GCP credentials
• Production secrets
• Session tokens

Example

Instead of submitting:

“Here is our AWS access key and secret. Fix this authentication error.”

the employee should remove the credentials and provide only the minimum non-sensitive technical information required for troubleshooting.

---

12. AI-Generated Content

AI-generated content may contain:

• Incorrect information
• Outdated information
• Fabricated references
• Biased conclusions
• Incomplete analysis
• Security weaknesses
• Copyright or licensing concerns

Users must review AI-generated content before relying on it.

For important business content, the reviewer should verify:

• Accuracy
• Relevance
• Completeness
• Confidentiality
• Security
• Legal/compliance implications
• Source credibility

---

13. Human Review

AI must not replace appropriate human judgment for high-impact activities.

Human review should be performed before AI output is used for significant:

• Customer decisions
• Security decisions
• Legal conclusions
• Financial decisions
• Employment decisions
• Compliance conclusions
• Risk assessments
• Production changes
• Contractual commitments

The level of review should be proportionate to the risk.

---

14. AI and Customer Communications

AI-generated customer communications should be reviewed before being sent where the content could materially affect:

• Customer commitments
• Security statements
• Contractual obligations
• Pricing
• Legal positions
• Incident communications
• Regulatory communications
• Service availability
• Technical instructions

Employees remain responsible for the accuracy of communications they send.

---

15. AI and Security Operations

AI may be used to support security activities such as:

• Log analysis
• Alert summarization
• Threat intelligence analysis
• Security documentation
• Detection-rule development
• Security research
• Code review
• Incident analysis

However:

• Sensitive information must be handled appropriately.
• AI output must be validated.
• Security decisions should not rely blindly on AI output.
• High-impact actions require appropriate human authorization.

---

16. AI for Incident Response

During a security incident, employees must not upload sensitive incident information to public AI tools unless specifically authorized.

Potentially sensitive information may include:

• Customer information
• Attack details
• Credentials
• Logs containing personal data
• Vulnerability details
• Security architecture
• Internal IP addresses
• Incident response records

AI may be used through an approved enterprise service where the organization’s security and privacy requirements are satisfied.

---

17. AI and Confidential Business Information

Confidential business information should only be processed by AI services that are approved for the relevant classification.

Examples include:

• Business plans
• Financial forecasts
• Pricing strategy
• Contracts
• M&A information
• Customer proposals
• Internal audit information
• Security assessments
• Risk registers
• Intellectual property

When in doubt:

Do not upload the information → Consult Security/IT → Confirm whether the AI tool is approved.

---

18. AI Output and Intellectual Property

Users should consider intellectual property and licensing implications when using AI-generated content or code.

Users should:

• Review applicable tool/provider terms.
• Avoid assuming that AI-generated material is automatically free from third-party rights.
• Review generated code and dependencies.
• Follow organizational intellectual-property requirements.
• Obtain appropriate review before externally distributing AI-generated material where necessary.

---

19. AI-Generated Code and Software Dependencies

Where AI generates software code, developers should verify:

• Code quality
• Security
• Licensing
• Dependency provenance
• Vulnerabilities
• Secrets
• Malicious or unsafe functionality
• Compatibility
• Test coverage

AI-generated code should pass the organization’s normal development and security controls.

AI should not become a mechanism for bypassing:

• Code review
• SAST
• SCA
• DAST
• Security testing
• VAPT
• Change management
• Production approval

---

20. AI and Decision Making

AI should generally be treated as a decision-support mechanism rather than an autonomous authority for significant organizational decisions.

Users should understand:

• What information the AI used.
• Whether the information is reliable.
• Whether the output can be independently verified.
• What limitations apply.
• What human approval is required.

The organization should identify higher-risk AI use cases that require additional governance.

---

21. AI-Generated Images, Audio, and Video

Users must ensure that AI-generated or AI-modified media is used appropriately.

Users must not use AI to:

• Impersonate individuals without authorization.
• Create fraudulent representations.
• Misrepresent events or statements.
• Violate privacy.
• Infringe applicable intellectual-property rights.
• Produce prohibited or illegal material.

Where disclosure of AI-generated or AI-modified content is appropriate or required, users should provide suitable disclosure.

---

22. AI Accounts and Access

AI services used for business purposes should use approved organizational accounts where available.

Users should:

• Use organizational authentication.
• Enable MFA where supported.
• Avoid sharing AI accounts.
• Protect API keys.
• Follow role-based access requirements.
• Remove access when no longer required.

Business AI accounts should not be created using personal email addresses where organizational accounts are available and required.

---

23. AI API Keys and Integrations

AI API keys must be treated as confidential credentials.

They should:

• Be stored in approved secrets-management systems.
• Never be committed to source code.
• Not be shared through chat or email.
• Be rotated when compromised.
• Have appropriate permissions and usage limits.
• Be monitored where appropriate.

---

24. Third-Party AI Services

Before an AI service is approved for business use, the organization should consider:

• Provider security
• Data processing
• Data retention
• Model training/use of submitted data
• Data location
• Subprocessors
• Encryption
• Access control
• Incident notification
• Availability
• Contractual terms
• Privacy requirements
• Exit and deletion arrangements

The level of review should be proportionate to the information and business process involved.

---

25. AI Vendor and Supplier Risk

Where an AI service processes confidential, customer, personal, or otherwise sensitive information, the supplier may need to undergo appropriate security/privacy assessment.

Possible evidence includes:

• Security certifications/assurance
• Security questionnaire
• Privacy documentation
• Data processing agreement
• Contract
• Data retention information
• Subprocessor information
• Incident response commitments
• Business continuity information

The AI service should be recorded in the organization’s SaaS Application Register where applicable.

---

26. Prompt Security

Users should treat prompts as business information where they contain sensitive content.

Users should:

• Provide only necessary information.
• Remove unnecessary identifiers.
• Avoid secrets.
• Avoid confidential information unless approved.
• Avoid copying entire documents when only a small section is required.
• Review the output before reuse.

Example

Instead of uploading an entire customer contract, provide only the specific non-sensitive clause requiring analysis where authorized and appropriate.

---

27. Prompt Injection and Untrusted AI Content

Users and developers should recognize that AI systems may process untrusted instructions embedded in:

• Documents
• Websites
• Emails
• Code
• Customer content
• External data
• Uploaded files

AI output should not automatically be trusted merely because it was generated by an AI system.

Where AI is integrated into applications, developers should consider risks such as:

• Prompt injection
• Data leakage
• Excessive agency
• Unauthorized tool use
• Insecure output handling
• Model manipulation
• Sensitive information disclosure

---

28. AI-Integrated Applications

Applications that use AI should follow applicable software security requirements.

Before deployment, consider:

• AI model/provider
• Data flows
• Input data
• Output data
• Access controls
• Prompt security
• API security
• Logging
• Monitoring
• Data retention
• Third-party dependencies
• Security testing
• Failure handling
• Human oversight

AI functionality should be included in relevant security architecture and risk assessments.

---

29. AI Risk Assessment

Significant AI use cases should be assessed for relevant risks.

Consider:

• Confidentiality
• Integrity
• Availability
• Privacy
• Accuracy
• Bias
• Intellectual property
• Security
• Regulatory requirements
• Customer impact
• Supplier dependency
• Operational dependency

Example:

AI customer-support assistant

Potential risks:

• Customer data exposure
• Incorrect response
• Unauthorized actions
• Prompt injection
• Excessive system permissions
• Provider outage

Treatment may include:

• Data minimization
• Access restrictions
• Human review
• Approved provider
• Logging
• Testing
• Output validation
• Restricted tool permissions

---

30. AI Use Cases

The organization may categorize AI use cases.

Low-Risk Examples

• Public-content summarization
• Brainstorming
• Translation of public information
• Formatting
• Drafting generic content

Moderate-Risk Examples

• Internal document summarization
• Code assistance
• Security analysis
• Internal research
• Business analysis

Higher-Risk Examples

• Customer decisions
• Employee decisions
• Processing sensitive personal information
• Automated security actions
• AI agents with production access
• AI making financial/legal/compliance decisions
• AI directly controlling business-critical systems

Higher-risk use cases should receive additional review and approval.

---

31. AI Agents and Autonomous Actions

AI agents that can perform actions in organizational systems require additional controls.

Examples include agents that can:

• Send emails
• Modify tickets
• Execute code
• Access databases
• Modify cloud resources
• Create users
• Change configurations
• Purchase services

Such systems should use:

• Least privilege
• Restricted permissions
• Approved tools
• Authentication
• Logging
• Monitoring
• Human approval for high-impact actions
• Rate limits
• Defined boundaries
• Emergency disablement mechanisms

An AI agent should not receive unrestricted administrative access merely because it is technically capable of using it.

---

32. Monitoring AI Usage

Where appropriate and legally permissible, the organization may monitor business AI usage to:

• Detect unauthorized AI services.
• Identify sensitive-data exposure.
• Monitor approved AI use.
• Investigate incidents.
• Assess compliance.
• Improve AI governance.

Monitoring should be proportionate and consider applicable privacy requirements.

---

33. Reporting AI Security Incidents

Users must report suspected AI-related incidents.

Examples include:

• Confidential information sent to an unauthorized AI service.
• AI account compromise.
• API key exposure.
• Unexpected AI data disclosure.
• AI-generated malicious content.
• Prompt injection.
• Unauthorized AI agent action.
• Incorrect AI output causing significant business impact.
• Unauthorized AI tool usage.

Response

Detect → Stop/Contain → Report → Assess → Investigate → Remediate → Verify → Learn

---

34. Exceptions

Exceptions to this policy must be:

• Business justified.
• Risk assessed.
• Approved by an authorized person.
• Documented.
• Time-bound where appropriate.
• Reviewed periodically.

An exception should record:

• AI service/use case
• Data involved
• Business reason
• Risk
• Compensating controls
• Approver
• Expiry/review date

---

35. Employee Responsibilities

Employees must:

• Use only approved AI tools for business information.
• Protect confidential information.
• Avoid entering secrets into AI systems.
• Review AI output.
• Follow data-classification requirements.
• Follow security and privacy requirements.
• Report AI-related incidents.
• Complete required AI/security awareness training.

Employees remain accountable for how AI-generated output is used.

---

36. Management Responsibilities

Management should:

• Approve appropriate AI use.
• Provide resources for secure AI adoption.
• Define risk tolerance for AI use.
• Ensure significant AI risks are addressed.
• Support employee awareness.
• Review significant AI-related incidents and risks.

---

37. IT and Information Security Responsibilities

IT/Security should:

• Maintain approved AI tools.
• Assess AI security risks.
• Support secure configuration.
• Monitor relevant AI usage.
• Review AI suppliers where appropriate.
• Establish security requirements.
• Support incident response.
• Periodically review AI risks.

---

38. Legal/Privacy Responsibilities

Where applicable, Legal/Privacy should support:

• Privacy assessments
• Data processing requirements
• Contract review
• Intellectual-property considerations
• Regulatory requirements
• Customer contractual requirements
• Cross-border data-transfer considerations

The level of legal/privacy review should be proportionate to the AI use case.

---

39. Approved AI Tool Register

The organization should maintain an AI Tool Register where appropriate.

ID	AI Tool	Provider	Purpose	Data Allowed	Owner	Security Review	Privacy Review	Status
AI-001	Enterprise AI Assistant	Provider	Productivity	Internal	IT	Completed	Completed	Approved
AI-002	Coding Assistant	Provider	Development	Approved source code	Engineering	Completed	As applicable	Approved
AI-003	Public AI Tool	Provider	General research	Public only	Security	Restricted	N/A	Restricted

The actual tools and approval status should be based on the organization’s own assessment.

---

40. AI Use Case Register

For significant AI applications, maintain an AI Use Case Register.

Field	Example
AI Use Case ID	AIUC-001
Use Case	Customer Support Assistant
Business Owner	Customer Support
AI Provider	Approved Provider
Purpose	Support response assistance
Data Processed	Customer support information
Classification	Confidential
Personal Data	Yes, where applicable
Human Review	Required
Risk Level	High
Security Review	Completed
Privacy Review	Completed
Supplier Review	Completed
Related Risk	AI-R-001
Status	Approved

---

41. AWS SaaS Startup Example

Consider a SaaS company using AI to help its customer-support team draft responses.

Data Flow

Customer Ticket

↓

Support Platform

↓

Approved AI Service

↓

Draft Response

↓

Human Support Agent

↓

Customer

Required Controls

• Approved AI provider
• Defined data scope
• Data minimization
• Access control
• Contractual review
• Privacy assessment where applicable
• Prompt restrictions
• Human review
• Logging
• Incident reporting
• Periodic risk review

The AI should assist the support employee rather than automatically sending unrestricted responses to customers unless the use case has been specifically assessed and approved.

---

42. Audit Evidence

Possible evidence includes:

• AI Acceptable Use Policy
• Approved AI Tool Register
• AI Use Case Register
• AI risk assessments
• Security assessments
• Privacy assessments
• Supplier assessments
• Contracts
• Data-processing agreements
• Access-control records
• AI awareness training
• Security monitoring
• Incident records
• Exception approvals
• AI application architecture
• Security testing
• Human-review records where required

An auditor may sample an AI use case and trace:

Use Case → Data → AI Provider → Risk → Controls → Approval → Operation → Monitoring → Review

---

43. Common Mistakes

Mistake 1 — Simply banning AI

A complete prohibition may push employees toward unauthorized “shadow AI” usage.

A controlled approach is generally more practical:

Identify → Assess → Approve → Control → Monitor → Review

Mistake 2 — Allowing public AI for confidential information

Employees may unintentionally expose customer or company information.

Mistake 3 — Assuming AI output is correct

AI output can be inaccurate or incomplete.

Mistake 4 — Ignoring AI coding tools

Source code, secrets, dependencies, and intellectual-property considerations require specific controls.

Mistake 5 — Ignoring AI agents

AI systems with the ability to perform actions require stronger access and monitoring controls.

Mistake 6 — No approved-tool list

Employees may not know which AI services are permitted.

Mistake 7 — No human accountability

AI-generated output should not automatically become an organizational decision.

Mistake 8 — No supplier assessment

AI services may process organizational information outside the organization’s direct environment.

---

44. Startup-Friendly AI Governance Model

A startup does not necessarily need a complex AI governance platform.

A practical initial model can use:

1. Approved AI Tool Register

What AI tools can employees use?

2. AI Use Case Register

How is the organization using AI?

3. AI Risk Assessment

What could go wrong?

4. Data Rules

What information can and cannot be entered?

5. Human Review

Which AI outputs require verification or approval?

6. Supplier Review

Who processes the information?

7. Incident Process

What happens if AI causes or contributes to a security incident?

This provides a practical foundation without requiring a large AI governance platform.

---

45. Relationship with Other ISMS Documents

The AI Acceptable Use Policy should connect with:

Information Security Policy
→ Overall security requirements

Acceptable Use Policy
→ General technology usage

Employee IT Usage Policy
→ Employee IT responsibilities

Information & Asset Inventory
→ AI-related systems and information

Data Inventory
→ Information processed by AI

Asset Classification Procedure
→ Classification and handling requirements

Access Control Policy
→ AI accounts and permissions

SaaS Application Register
→ Third-party AI services

Supplier Security Assessment
→ AI provider risk

Risk Assessment
→ AI-specific risks

Incident Management
→ AI-related incidents

Security Awareness Policy
→ Employee AI awareness

Secure Development Policy
→ AI-generated code and AI-enabled applications

Privacy/RoPA Processes
→ Personal-data processing through AI

The relationship is:

AI Use Case → Data → Provider/System → Risk → Approval → Controls → Human Review → Monitoring → Incident Response → Review

---

46. Quick Audit Checklist

Check	Yes/No	Evidence
Is an AI Acceptable Use Policy approved?
Are approved AI tools identified?
Are prohibited AI uses defined?
Are confidential and restricted data rules defined?
Are customer data requirements addressed?
Are personal-data requirements addressed?
Are passwords, keys, and credentials prohibited from unauthorized AI use?
Are source-code requirements defined?
Is AI-generated output subject to appropriate review?
Are AI suppliers assessed where required?
Are AI use cases risk assessed where appropriate?
Are AI agents addressed?
Are AI API keys protected?
Are AI-related incidents reportable?
Are exceptions controlled?
Are employees trained?
Is AI usage monitored where appropriate?
Are AI tools periodically reviewed?
Is the policy periodically reviewed?

---

47. Policy Review

This policy should be reviewed periodically and whenever significant changes occur, including:

• Introduction of new AI technologies.
• Adoption of new AI providers.
• Significant changes in AI capabilities.
• New AI-enabled business processes.
• Security incidents involving AI.
• Changes in applicable legal or regulatory requirements.
• Changes to customer contractual requirements.
• Significant changes in the organization’s risk profile.

The policy should be updated when existing controls no longer adequately address the organization’s AI risks.

---

48. Final Principle

The objective of AI governance is not simply to prevent employees from using AI.

It is to enable useful AI adoption while protecting the organization’s information, customers, employees, systems, and reputation.

The practical model is:

Identify → Assess → Approve → Minimize Data → Use Securely → Validate → Monitor → Report → Improve

Or, more simply:

Use AI where it creates value. Protect the data. Verify the output. Keep humans accountable.