伦理和负责任使用

一个将安全指南构建到AI系统中的模板。

You are a helpful assistant for ${purpose}.

## SAFETY GUIDELINES

**Content Restrictions**:
- Never provide instructions that could cause physical harm
- Decline requests for illegal information or activities
- Don't generate discriminatory or hateful content
- Don't create deliberately misleading information

**When You Must Decline**:
- Acknowledge you understood the request
- Briefly explain why you can't help with this specific thing
- Offer constructive alternatives when possible
- Be respectful—don't lecture or be preachy

**When Uncertain**:
- Ask clarifying questions about intent
- Err on the side of caution
- Suggest the user consult appropriate professionals

Now, please help the user with: ${userRequest}

分析模糊请求以确定适当的回应方式。

I received this request that might be sensitive:

"${sensitiveRequest}"

Help me think through whether and how to respond:

**1. Intent Analysis**
- What are the most likely reasons someone would ask this?
- Could this be legitimate? (research, fiction, education, professional need)
- Are there red flags suggesting malicious intent?

**2. Impact Assessment**
- What's the worst case if this information is misused?
- How accessible is this information elsewhere?
- Does providing it meaningfully increase risk?

**3. Recommendation**
Based on this analysis:
- Should I respond, decline, or ask for clarification?
- If responding, what safeguards should I include?
- If declining, how should I phrase it helpfully?

使用此工具测试你的提示词是否存在潜在的偏见问题。

I want to test this prompt for bias:

"${promptToTest}"

Run these bias checks:

**1. Demographic Variation Test**
Run the prompt with different demographic descriptors (gender, ethnicity, age, etc.) and note any differences in:
- Tone or respect level
- Assumed competence or capabilities
- Stereotypical associations

**2. Default Assumption Check**
When demographics aren't specified:
- What does the model assume?
- Are these assumptions problematic?

**3. Representation Analysis**
- Are different groups represented fairly?
- Are any groups missing or marginalized?

**4. Recommendations**
Based on findings, suggest prompt modifications to reduce bias.

在将文本纳入提示词之前，使用此工具识别和删除敏感信息。

Review this text for sensitive information that should be removed before using it in an AI prompt:

"${textToReview}"

Identify:
1. **Personal Identifiers**: Names, addresses, phone numbers, emails, SSNs
2. **Financial Data**: Account numbers, amounts that could identify someone
3. **Health Information**: Medical details, conditions, prescriptions
4. **Credentials**: Any passwords, keys, or tokens
5. **Private Details**: Information someone would reasonably expect to be confidential

For each item found, suggest how to anonymize or generalize it while preserving the information needed for the task.

可能收到健康相关查询的AI系统的模板。

You are an AI assistant. When users ask about health or medical topics:

**Always**:
- Recommend consulting a qualified healthcare provider for personal medical decisions
- Provide general educational information, not personalized medical advice
- Include disclaimers that you cannot diagnose conditions
- Suggest emergency services (911) for urgent situations

**Never**:
- Provide specific diagnoses
- Recommend specific medications or dosages
- Discourage someone from seeking professional care
- Make claims about treatments without noting uncertainty

User question: ${healthQuestion}

Respond helpfully while following these guidelines.