What Check Point AI Guardrails does
Check Point AI Guardrails, formerly Lakera Guard, is a runtime security service for prompt attacks, data leakage, harmful content, malicious links, and off-policy agent actions.
Lakera Guard is now documented within Check Point AI Security as AI Guardrails. The runtime layer can screen direct and indirect prompt injection, jailbreaks, sensitive-data exposure, prohibited content, links, and agent tool behavior. Broader AI Agent Security adds discovery and risk assessment across supported platforms. The product can sit around model calls without replacing the application's authentication or business rules.
Official documentation offers a free starting account, while production entitlements and full AI Agent Security are plan- or contract-dependent. Total cost depends on screening volume, content length, detectors, latency requirements, logging, environments, support, and deployment arrangement. Ask for exact request units, overages, retention, regional processing, SIEM export, service levels, and acquisition-era contract identity.
No detector catches every novel attack, and strict policies can block legitimate text. Prompt logging is enabled by default for the SaaS dashboard according to current docs, though it can be disabled; that setting is material for confidential inputs. Run in observe mode, measure false positives and false negatives on your traffic, minimize metadata, protect keys, screen every relevant agent step, and keep allowlists, authorization, confirmations, sandboxing, and human escalation outside the model.
How Check Point AI Guardrails works
An application creates a project and policy, then sends the current interaction to the Guard API, including clean user input, model output, retrieved content, or agent tool messages as appropriate. Managed and custom detectors examine prompt attacks, PII or other leakage, content violations, malicious links, and off-task or disallowed tool behavior. The API returns a flagged decision and optional detector detail; the application decides whether to block, mask, warn, log, or escalate. Teams must tune thresholds and retain deterministic authorization around real actions.
How to set up Check Point AI Guardrails
Map the threat model
Identify users, documents, tools, secrets, regulated data, abuse cases, consequences, latency budget, and the controls that remain outside AI.
Create isolated projects
Separate applications and environments, issue scoped keys, choose data region and retention, and decide whether prompt logging is permitted.
Build a tailored policy
Enable relevant prompt, leakage, content, link, and agent-action defenses; add custom rules without treating the default policy as production-ready.
Observe and evaluate
Screen representative normal and adversarial traffic without blocking, label results, tune thresholds, test languages, and measure latency and misses.
Enforce in layers
Block or mask validated classes, export alerts, preserve deterministic tool authorization and confirmation, monitor drift, and rehearse bypass-safe outages.
Check Point AI Guardrails FAQs
Is Lakera Guard still available?
The product is now documented as Check Point AI Guardrails after Check Point's acquisition; Lakera API domains remain in current official documentation.
What can AI Guardrails detect?
Configured defenses cover prompt attacks, data leakage, content violations, malicious links, and selected off-policy or disallowed agent tool behavior.
Does it block requests automatically?
The API returns detection results. The integrating application chooses whether to block, mask, warn, log, or escalate.
Are prompts stored?
Current SaaS documentation says prompt and output logging is on by default for the dashboard and can be disabled. Verify retention and contract terms.
Can guardrails make an agent safe by themselves?
No. Use least-privilege tools, deterministic authorization, validation, sandboxing, rate limits, human confirmation, monitoring, and incident response as well.
Listing reviewed 2026-07-28. Product details and pricing can change; verify important terms on the provider's website.
Related Coding AI tools
Related AI guides
Reviews
Tell the community what you made, what worked, and what you wish you knew before starting.