実践におけるLLMのガードレール：本当に効果的なものとは

import re

class PromptSanitizer:
    def __init__(self):
        self.dangerous_patterns = [
            r"ignore\s+previous\s+instructions",
            r"system\s+prompt",
            r"you\s+are\s+now\s+free",
            r"break\s+out\s+of",
        ]

    def sanitize(self, prompt: str) -> str:
        for pattern in self.dangerous_patterns:
            prompt = re.sub(pattern, "[REDACTED]", prompt, flags=re.IGNORECASE)
        return prompt

class InputValidator:
    def __init__(self, max_length: int = 10000):
        self.max_length = max_length

    def validate(self, prompt: str) -> tuple[bool, str]:
        if len(prompt) > self.max_length:
            return False, f"Input too long: {len(prompt)} > {self.max_length}"
        return True, "OK"

class ContentFilter:
    def __init__(self):
        self.blocked_topics = [
            "violence", "hate speech", "self-harm",
            "sexual content", "illegal activities",
        ]

    def filter(self, prompt: str) -> tuple[bool, str]:
        prompt_lower = prompt.lower()
        for topic in self.blocked_topics:
            if topic in prompt_lower:
                return False, f"Blocked: {topic}"
        return True, "OK"

class ResponseValidator:
    def __init__(self):
        self.required_fields = ["answer", "confidence"]

    def validate(self, response: dict) -> tuple[bool, str]:
        for field in self.required_fields:
            if field not in response:
                return False, f"Missing field: {field}"
        return True, "OK"

class OutputFilter:
    def __init__(self):
        self.blocked_patterns = [
            r"kill\s+someone",
            r"bomb\s+recipe",
            r"hate\s+speech",
            r"self-harm",
        ]

    def filter(self, response: str) -> tuple[bool, str]:
        for pattern in self.blocked_patterns:
            if re.search(pattern, response, re.IGNORECASE):
                return False, f"Blocked: {pattern}"
        return True, "OK"

class FactChecker:
    def __init__(self):
        self.known_facts = {
            "capital of france": "Paris",
            "population of usa": "330 million",
            "speed of light": "299,792,458 m/s",
        }

    def check(self, claim: str) -> tuple[bool, str]:
        claim_lower = claim.lower()
        for fact, truth in self.known_facts.items():
            if fact in claim_lower and truth not in claim_lower:
                return False, f"Fact check failed: {fact}"
        return True, "OK"

import time
from collections import deque

class RateLimiter:
    def __init__(self, max_requests: int = 10, window: int = 60):
        self.max_requests = max_requests
        self.window = window
        self.requests = deque()

    def allow(self) -> bool:
        now = time.time()
        while self.requests and self.requests[0] < now - self.window:
            self.requests.popleft()

        if len(self.requests) >= self.max_requests:
            return False

        self.requests.append(now)
        return True

class TokenBudget:
    def __init__(self, max_tokens: int = 1000):
        self.max_tokens = max_tokens

    def validate(self, response: str) -> tuple[bool, str]:
        token_count = len(response.split())
        if token_count > self.max_tokens:
            return False, f"Token limit exceeded: {token_count} > {self.max_tokens}"
        return True, "OK"

class ContextManager:
    def __init__(self, max_context: int = 4096):
        self.max_context = max_context
        self.context = []

    def add(self, message: str):
        self.context.append(message)
        self.trim()

    def trim(self):
        while len(" ".join(self.context)) > self.max_context:
            self.context.pop(0)

class DataResidency:
    def __init__(self, allowed_regions: list[str]):
        self.allowed_regions = allowed_regions

    def validate(self, region: str) -> tuple[bool, str]:
        if region not in self.allowed_regions:
            return False, f"Region not allowed: {region}"
        return True, "OK"

import json
from datetime import datetime

class AuditLogger:
    def __init__(self, log_file: str = "audit.log"):
        self.log_file = log_file

    def log(self, request: dict, response: dict):
        entry = {
            "timestamp": datetime.now().isoformat(),
            "request": request,
            "response": response,
        }
        with open(self.log_file, "a") as f:
            f.write(json.dumps(entry) + "\n")

class SimpleGuardrails:
    def __init__(self):
        self.input_validator = InputValidator(max_length=10000)
        self.output_filter = OutputFilter()

    def process(self, prompt: str) -> str:
        valid, message = self.input_validator.validate(prompt)
        if not valid:
            return f"Error: {message}"

        response = self.call_model(prompt)

        valid, message = self.output_filter.filter(response)
        if not valid:
            return f"Error: {message}"

        return response

class AdvancedGuardrails:
    def __init__(self):
        self.sanitizer = PromptSanitizer()
        self.input_validator = InputValidator(max_length=10000)
        self.content_filter = ContentFilter()
        self.output_filter = OutputFilter()
        self.rate_limiter = RateLimiter(max_requests=10)
        self.token_budget = TokenBudget(max_tokens=1000)

    def process(self, prompt: str) -> str:
        prompt = self.sanitizer.sanitize(prompt)

        valid, message = self.input_validator.validate(prompt)
        if not valid:
            return f"Error: {message}"

        valid, message = self.content_filter.filter(prompt)
        if not valid:
            return f"Error: {message}"

        if not self.rate_limiter.allow():
            return "Error: Rate limit exceeded"

        response = self.call_model(prompt)

        valid, message = self.output_filter.filter(response)
        if not valid:
            return f"Error: {message}"

        valid, message = self.token_budget.validate(response)
        if not valid:
            return f"Error: {message}"

        return response