What is Lakera?
Lakera Guard is now sold by Check Point as AI Guardrails, part of its AI Defense Plane suite. Its docs say it detects prompt attacks, data leakage, content violations and off-policy agent behavior in real time through the Guard API, and it runs as SaaS or self-hosted.
Who it's for
Security teams and organizations building or deploying LLM apps and AI agents, according to the vendor's docs. The Community tier is meant for evaluation at low volume, while larger volumes, self-hosting and extra controls are listed for Enterprise.
Pros & cons
Strengths
- The docs say AI Guardrails works with any model or provider, including hosted (OpenAI, Anthropic, Cohere), open-source and custom or fine-tuned models.
- The docs say AI Agent Security discovers agents, their tools and connected MCP servers, rates risk per agent, and applies guardrails to prompts, outputs, tool calls, tool responses and tool descriptions.
- The data leakage docs list built-in detectors (names, US addresses, US phone numbers, emails, IP addresses, credit card numbers, IBANs, US Social Security numbers) plus custom guardrails, trigger words and custom entity types, with a choice to block or mask.
- Check Point's privacy data sheet lists customer-selected hosting in the EU, U.S. or Singapore for the cloud service.
- Self-hosting is documented via Helm on Kubernetes, Docker or another OCI runtime, or air-gapped; the docs say it is meant to keep all data in the customer's infrastructure.
- The docs say Community accounts get 10k screening requests a month at no stated charge, which allows testing before talking to sales.
Limitations
- Community is limited to 10k screening requests a month, per the official docs. The same page lists 1MB of context per request, RBAC for the Dashboard, SIEM export to an S3-compatible bucket and data retention control as Enterprise features.
- Self-hosting requires an AI Guardrails Enterprise license, and the docs say the Dashboard is not currently available to self-hosting customers.
- No prices are published. A June 2026 third-party review says Lakera does not publish list prices and that new paid quotes go through Check Point enterprise sales (Reported).
- Check Point's data sheet says cloud data is retained for the subscription plus three months after termination; customers can delete prompts or restrict retention of pseudonymized prompts. The data sheet does not address training on customer data, though the data leakage docs say Check Point's AI models will not be trained on PII.
- The vendor's docs do not describe the detection techniques or models behind the guardrails. Earlier third-party write-ups mention an 8,000-token prompt limit on Community and a possible Pro tier; neither could be confirmed on an official page (Reported).
Plans and pricing
- Community: $0 (Reported). Up to 10k screening requests a month per the official docs; the $0 price comes from third-party pages. Aimed at evaluation and low-volume use.
- Enterprise: Contact sales. Flexible monthly request package, up to 1MB context per request, Dashboard RBAC, SIEM log export, retention control; self-hosting needs an Enterprise license.
The official pricing page could not be fetched, so plan details come from the official docs (Community limit, Enterprise features) and third-party pages (the $0 price). Enterprise pricing is not published and quotes go through Check Point sales. Confirm on the live pricing page.
Details checked October 2026 against the vendor's own pages. Prices and limits change, so confirm before you buy. Official pricing page ↗
Sources
This profile was compiled from these public pages. We have not run hands-on tests.
Lakera — at a glance
Lakera Guard is now sold by Check Point as AI Guardrails, part of its AI Defense Plane suite. Its docs say it detects prompt attacks, data leakage, content violations and off-policy agent behavior in real time through the Guard API, and it runs as SaaS or self-hosted.
Try Lakera →