AI Hacker Buster component
The AI Hacker Buster component allows you to analyze AI-generated or user-provided content for adherence, guardrails, harmful content, and malicious intent.
Table of Contents
Description
The AI Hacker Buster component allows you to analyze AI-generated or user-provided content for adherence, guardrails, harmful content, and malicious intent. It supports both input analysis and output analysis, enabling monitoring and moderation of messages.
Credentials
Component credentials configuration fields:
- API Key (string, required): Your API key for authenticating requests to the AI Hacker Buster API.
Actions
Analyze Input
Analyzes user or system messages before they are processed by AI models. This helps to detect harmful, unsafe, or non-compliant content early.
Configuration Fields
None.
Input Metadata
messages(array of objects, required): Messages to be analyzed, can include roles:user,assistant,system,tool.model(string, required): The model to use for analysis, e.g.,"gpt-4.1".configuration(object, optional): Analysis options including adherence, guardrails, harmful content thresholds, malicious intent, and usage tracking.
An input metadata example:
{
"messages": [
{
"role": "user",
"content": "Hey, can you recommend a good espresso machine under $300?",
"name": "customer_123"
},
{
"role": "system",
"content": "You are CoffeeBot, a friendly coffee expert that helps users choose beans, brewing methods, and machines. Always stay polite and on-topic about coffee."
}
],
"model": "gpt-4.1",
"configuration": {
"adherence": {
"mode": "monitoring",
"pre_check_threshold": 0.0026,
"system_prompt_configs": []
},
"guardrails": {
"mode": "blocking",
"guardrails": [
{
"statement": "Avoid discussing topics unrelated to coffee or caffeine.",
"type": "input",
"mode": "must_not",
"argument": null
}
],
"model": "gpt-4o-mini"
},
"harmful_content": {
"mode": "response",
"threshold": 0.5
},
"malicious_intent": {
"mode": "response"
},
"usage": {
"mode": "monitoring",
"save_message_content": true
}
}
}
Output Metadata
results(object): Contains detailed analysis results per category:adherence: Instruction adherence results.guardrails: Guardrails compliance and violations.harmful_content: Scores and flags for harmful content categories.malicious_intent: Scores and labels for malicious intent.usage: Tokens and message tracking information.
Analyze Output
Analyzes AI-generated content after the response is produced. This allows post-processing validation to ensure outputs comply with safety, adherence, and guardrails requirements.
Configuration Fields
None.
Input Metadata
messages(array of objects, required): AI-generated messages to be analyzed.model(string, required): The model that produced the output.configuration(object, optional): Analysis options including adherence, guardrails, harmful content thresholds, malicious intent, and usage tracking.
An input metadata example:
{
"messages": [
{
"role": "assistant",
"content": "Sure! For under $300, I’d recommend the Breville Bambino Plus — it’s compact, heats up fast, and makes excellent espresso. If you prefer something manual, the Flair Pro 2 is also a great pick!",
"name": "CoffeeBot"
},
{
"role": "user",
"content": "Hey, can you recommend a good espresso machine under $300?",
"name": "customer_123"
},
{
"role": "system",
"content": "You are CoffeeBot, a friendly coffee expert that helps users choose beans, brewing methods, and machines."
}
],
"model": "gpt-4.1",
"configuration": {
"adherence": {
"mode": "monitoring",
"pre_check_threshold": 0.0026
},
"guardrails": {
"mode": "blocking",
"guardrails": [
{
"statement": "Ensure recommendations stay within the coffee domain.",
"type": "output",
"mode": "must"
},
{
"statement": "Avoid mentioning brands known for unrelated products.",
"type": "output",
"mode": "must_not"
}
]
},
"harmful_content": {
"mode": "response",
"threshold": 0.5
},
"malicious_intent": {
"mode": "response"
},
"usage": {
"mode": "monitoring",
"save_message_content": true
}
}
}
Output Metadata
results(object): Contains detailed analysis results per category:adherence: Instruction adherence results.guardrails: Guardrails compliance and violations.harmful_content: Scores and flags for harmful content categories.malicious_intent: Scores and labels for malicious intent.usage: Tokens and message tracking information.