Skip to content
Fathom Layer

Versus Engine

Lakera Guard vs NVIDIA Garak

Recorded capabilities and trade-offs for Lakera Guard and NVIDIA Garak, side by side. Missing information is not treated as a tie.

Capabilities on record

Lakera Guard

  • Real-time request and response scanning
  • Single chat-completions-style API endpoint
  • Prompt injection and jailbreak detection
  • PII and data-leakage detection
  • Self-hosted deployment option

Capabilities on record

NVIDIA Garak

  • CLI scanner combining dynamic and static probes
  • Jailbreak, prompt-injection, toxicity, leakage and hallucination probes
  • Adapters for OpenAI, Anthropic, Hugging Face, Ollama and more
  • Machine-readable JSONL reports
  • Extensible probe and detector plugins
Lakera Guard

Lakera Guard

A real-time API that inspects prompts and model output for prompt injection, jailbreaks, system-prompt extraction and PII leakage, sitting between your application and any LLM.

Editorial notes on record

Recorded strengths

  • Model-agnostic — sits in front of hosted APIs or local models alike
  • Covers direct and indirect prompt injection, jailbreaks, system-prompt extraction and PII leakage
  • Filters are updated from a large, real-world attack dataset (Gandalf)

Recorded limitations

  • Closed-source commercial service — an external dependency on the request path
  • Detection-based: a high block rate is not a guarantee
  • Headline accuracy and latency numbers are vendor-published, not independently measured
NVIDIA Garak

NVIDIA Garak

An open-source command-line vulnerability scanner for LLMs — it probes a model for jailbreaks, prompt injection, data leakage, toxicity and forced hallucination, then writes a structured report of what got through.

Editorial notes on record

Recorded strengths

  • Fully open-source (Apache 2.0), no vendor lock-in
  • Works against hosted APIs, Hugging Face models and local endpoints alike
  • Large, actively maintained probe library with machine-readable output

Recorded limitations

  • Command-line tool — results need security literacy to act on
  • Scanning hosted models costs API tokens and can hit rate limits
  • Identifies weaknesses only; remediation is on you