AI Safety y Guardrails

Modelo de seguridad en profundidad

⚠ Advertencia

Input validation: filtrar antes del LLM

const BLOCKED_PATTERNS = [
  /ignore (all |previous |above )?instructions/i,
  /you are now/i,
  /pretend (to be|you are)/i,
  /disregard (your |all )?rules/i,
  /jailbreak/i,
  /DAN mode/i,
]

const MAX_INPUT_LENGTH = 10000
const MAX_MESSAGES = 50

export function validateInput(input: string, messages: any[]): { valid: boolean; error?: string } {
  // Length check
  if (input.length > MAX_INPUT_LENGTH) {
    return { valid: false, error: 'Input demasiado largo' }
  }

  // History check
  if (messages.length > MAX_MESSAGES) {
    return { valid: false, error: 'Conversación demasiado larga' }
  }

  // Injection patterns
  for (const pattern of BLOCKED_PATTERNS) {
    if (pattern.test(input)) {
      return { valid: false, error: 'Contenido no permitido' }
    }
  }

  return { valid: true }
}

Output filtering: validar respuestas

const PII_PATTERNS = {
  email: /[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}/g,
  phone: /\+?[0-9]{10,15}/g,
  ssn: /\d{3}-\d{2}-\d{4}/g,
  creditCard: /\d{4}[\s-]?\d{4}[\s-]?\d{4}[\s-]?\d{4}/g,
}

export function filterOutput(text: string): string {
  let filtered = text

  // Redact PII
  for (const [type, pattern] of Object.entries(PII_PATTERNS)) {
    filtered = filtered.replace(pattern, `[REDACTED_${type.toUpperCase()}]`)
  }

  return filtered
}

export function checkOutputSafety(text: string): boolean {
  const dangerousPatterns = [
    /how to (make|build|create) (a )?bomb/i,
    /instructions for (making |creating )?weapons/i,
  ]

  return !dangerousPatterns.some(p => p.test(text))
}

Rate limiting con Upstash

import { Ratelimit } from '@upstash/ratelimit'
import { Redis } from '@upstash/redis'

const ratelimit = new Ratelimit({
  redis: Redis.fromEnv(),
  limiter: Ratelimit.slidingWindow(10, '1 m'), // 10 requests per minute
  analytics: true,
})

export async function middleware(request: Request) {
  if (request.url.includes('/api/chat')) {
    const ip = request.headers.get('x-forwarded-for') ?? '127.0.0.1'
    const { success, limit, remaining } = await ratelimit.limit(ip)

    if (!success) {
      return new Response('Rate limit exceeded', {
        status: 429,
        headers: {
          'X-RateLimit-Limit': limit.toString(),
          'X-RateLimit-Remaining': remaining.toString(),
        }
      })
    }
  }
}

Tip Pro