> For the complete documentation index, see [llms.txt](https://charity.childsafetyfund.org/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://charity.childsafetyfund.org/policies/operations/ai.md).

# AI

This policy establishes how Child Safety Fund (CSF) may use artificial intelligence in ways that support its mission, protect children and families, and preserve public trust. It is intended to ensure that AI strengthens research, strategy, operations, and communication without displacing human judgment in matters involving child safety, family welfare, or community well-being.

***

### **Scope**

This policy applies to all CSF staff, contractors, consultants, volunteers, and partners who use AI on behalf of CSF. It applies to both third-party and internally configured AI tools.

***

### **Policy Position**

CSF uses AI as a support tool, not as a decision-maker.

AI may be used to:

* Synthesize research and publicly available information
* Support strategy development, planning, and documentation
* Analyze anonymized, aggregated, or community-level data
* Assist with administrative and operational tasks
* Improve communication, drafting, and knowledge organization

AI may not be used to:

* Make final decisions about child safety, family welfare, service eligibility, referrals, or interventions
* Process identifiable case information about children or families unless expressly approved under a separate written exception process and all legal, privacy, and security requirements are met
* Replace professional judgment in high-stakes contexts
* Generate or present unsupported conclusions as factual findings

***

### **Governing Principles**

CSF will govern AI use according to the following principles:

* Child safety first
* Human accountability at all times
* Purpose limitation and data minimization
* Transparency and explainability
* Fairness and non-discrimination
* Privacy and security by design
* Proportionate, risk-based use
* Continuous monitoring and improvement

***

### **Risk Tiers**

CSF will classify AI use into three tiers.

Low-risk use

* Internal drafting, summarization, formatting, brainstorming, and administrative support
* May proceed with standard staff training and approved tools

Moderate-risk use

* Research synthesis, strategy development, community-level analysis, policy drafting, communications planning, and non-case-specific program design
* Requires manager review and use of approved tools

High-risk use

* Any use related to child safety decisions, family welfare decisions, referrals, case assessments, eligibility determinations, or processing of identifiable child or family information
* Prohibited unless explicitly approved in writing by executive leadership under a documented exception process with legal, privacy, and operational safeguards

***

### **Approved Data Use**

CSF may use AI with:

* Publicly available research and policy information
* Internal operational information that does not contain protected case data
* Anonymized or aggregated data
* Community-level or population-level information used for strategy or planning

CSF will not input identifiable child or family case information into AI systems as a standard practice.

Where consent is relevant, CSF will obtain informed consent before collecting or using community or participant data in connection with AI-supported work.

***

### **Privacy and Security**

CSF will:

* Comply with applicable privacy, child protection, and data security laws and obligations
* Limit access to sensitive information based on role and need
* Use approved platforms and configurations for AI-enabled work
* Minimize the amount of data shared with AI tools
* Prefer anonymization, aggregation, or de-identification before processing
* Maintain secure storage and transmission practices for AI-related materials

***

### **Human Oversight**

Human review is required for all moderate-risk and high-risk AI use.

No AI output may be treated as final in matters affecting children, families, or communities without human review by an authorized professional.

Staff must:

* Verify important claims before use
* Check outputs for bias, errors, omissions, and overstatement
* Override AI outputs when professional judgment, evidence, or policy requires it
* Escalate questionable outputs when the consequences of error could be material

***

### **Transparency**

CSF will be transparent about how AI is used.

CSF will:

* Clearly document the role of AI in significant internal and external work products
* Label AI-generated content where disclosure is appropriate and useful
* Be prepared to explain, in plain language, how AI informed a recommendation, model, or document
* Publicly communicate its general AI practices through policy and governance documentation

***

### **Fairness and Bias Control**

CSF recognizes that AI systems may reproduce or amplify bias.

To reduce this risk, CSF will:

* Use diverse and representative source material where feasible
* Review outputs for harmful assumptions, exclusion, or disparate impact
* Include subject-matter experts and relevant stakeholders in reviewing high-consequence work
* Conduct periodic bias reviews of recurring AI-supported workflows
* Maintain clear standards for when human judgment must override AI recommendations

***

### **Third-Party Tools and Vendors**

CSF may use third-party AI tools only with appropriate oversight.

Before adoption or use at scale, CSF will assess:

* Data handling practices
* Privacy and security controls
* Transparency of the tool’s limitations and intended use
* Contract terms relevant to confidentiality, retention, and vendor responsibility
* Whether the tool is appropriate for the risk level of the intended use

CSF will not rely on vendor claims alone where risk is material.

***

### **Accountability**

Accountability for AI use is shared across the organization.

* Individual users are responsible for the outputs they rely on or circulate
* Managers are responsible for oversight of AI use within their teams or functions
* Executive leadership retains final accountability for organizational AI use and exception approvals
* CSF may designate a lightweight internal oversight function to maintain approved tools, review higher-risk uses, and coordinate policy updates

***

### **Training**

Before using AI on behalf of CSF, staff must receive baseline guidance on:

* Approved and prohibited uses
* Privacy and data handling expectations
* Human review requirements
* Bias and fairness risks
* Documentation and disclosure expectations

CSF will provide ongoing resources, examples, and updates as AI use evolves.

***

### **Documentation**

CSF will document significant AI-supported uses in proportion to their risk.

For moderate-risk and high-risk use cases, documentation should include:

* Purpose of use
* Tool used
* Data category used
* Reviewer or approver
* Key limitations or cautions
* Final human decision owner

***

### **Incidents and Violations**

Suspected misuse, policy violations, data exposure, or materially harmful AI outputs must be reported promptly to management.

CSF may suspend a tool, workflow, or user access while reviewing the issue. Confirmed violations may result in corrective action, retraining, process changes, vendor review, or discontinuation of the tool.

***

### **Feedback and Community Input**

CSF may gather feedback on AI use through:

* Staff feedback loops
* Partner and stakeholder input
* Community engagement processes where relevant
* Outcome reviews and implementation learning

When AI-supported work materially affects public-facing programs or communications, CSF should consider whether additional external input is warranted.

***

### **Review and Updates**

This policy will be reviewed periodically and updated as CSF’s AI use, legal obligations, risk profile, and operating practices evolve.

Any high-risk use not already authorized by this policy requires separate written review and approval before implementation.

***

### **Default Rule**

If there is uncertainty about whether a proposed AI use is appropriate, staff should pause use and seek review before proceeding.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://charity.childsafetyfund.org/policies/operations/ai.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
