Lakera Guard
Real-time API for detecting and preventing LLM prompt injections, jailbreaks, and other security threats. Provides content moderation and input validation for AI applications with low latency processing.
Overview
Lakera Guard is an AI security API that protects LLM applications against prompt injection, jailbreaks, and data leakage in real time. Acquired by Cisco in May 2025, the platform delivers 98%+ detection rates with sub-50ms latency across 100+ languages. Lakera's threat intelligence is powered by Gandalf, an adversarial AI game that has generated 80M+ attack prompts from over 1 million players. The API evaluates both inputs and outputs, returning allow/block/sanitize decisions that integrate with any LLM provider. The Zurich-based research team continues maintaining Guard's detection models under Cisco AI Defense.
The Verdict
Who Should Use Lakera Guard?
Best For
- Production LLM applications needing real-time protection
- Teams requiring low-latency security (<50ms)
- Applications handling sensitive data (PII, secrets)
- Multi-language deployments (100+ languages)
- Enterprise Cisco customers seeking AI security
Not Ideal For
- High-volume apps on tight budgets (usage-based pricing)
- Teams wanting self-hosted open-source (see LLM Guard)
- Simple prototypes or MVPs
- Organizations avoiding cloud dependencies
What's Great
- 98%+ detection rate for prompt injections and jailbreaks
- Sub-50ms latency for real-time protection
- Gandalf-powered threat intelligence (80M+ adversarial prompts)
- Works with any LLM provider (OpenAI, Claude, Llama, etc.)
- Comprehensive coverage: PII, secrets, toxic content, policy violations
- SOC 2 compliant with GDPR-aligned data handling
- Free Community tier for development and testing
Watch Out For
- Usage-based pricing can scale with volume
- Cloud-only unless on Enterprise plan (self-host option)
- Now part of Cisco (procurement through enterprise sales)
- Community tier limited to 10K requests/month
Pricing
View all features & details
Threat Detection
- Direct prompt injection
- Indirect prompt injection
- Jailbreak attempts
- System prompt extraction
- PII/secrets detection
- Toxic content filtering
Data Protection
- PII redaction
- Secret sanitization
- Data loss prevention
- Content moderation
- Policy enforcement
- Audit logging
Deployment
- Cloud SaaS (default)
- Self-hosted (Enterprise)
- Regional hosting options
- API-first integration
- SDK support
- Webhook notifications
Compliance
- SOC 2 certified
- GDPR compliant
- DPA available
- Data residency options
- HIPAA/PCI consideration
- Cisco AI Defense integration
How It Compares
| Feature | Lakera Guard | LLM Guard | NeMo Guardrails | Prompt Security |
|---|---|---|---|---|
| Type | Cloud API | Open Source | Open Source | Cloud API |
| Detection Rate | 98%+ | 90%+ | Custom | 95%+ |
| Latency | <50ms | Self-hosted | Self-hosted | ~100ms |
| Languages | 100+ | ~10 | Custom | 50+ |
| Free Tier | 10K req/mo | Unlimited | Unlimited | Limited |
| Threat Intel | Gandalf (80M+) | Community | None | Proprietary |
| Best For | Production APIs | Self-hosted | Custom flows | Enterprise |