Security for applications that take untrusted input.
Lakera Guard sits in front of LLM applications detecting prompt injection\, jailbreak attempts and data leakage — trained partly on attacks collected through Gandalf\, its public prompt-injection game played millions of times.
Using a game to crowdsource real attack data is a genuinely clever way to stay current with techniques that evolve weekly. Any application accepting untrusted input needs something here. Its pricing page served no details when this was written.
Quick Information
Platform
Web
Pricing
Free tier + Paid
API
Available
Category
Analytics
Pros and Cons
Pros
- Purpose-built for prompt injection defence
- Trained on real crowdsourced attacks
- Low-latency inline guardrails
- Detects data leakage too
- Strong security research reputation
Cons
- Pricing not published
- Adds latency to requests
- No guardrail catches everything
- Needs tuning to avoid false positives