Lakera is an AI security platform built for prompt injection protection and runtime guardrails in generative AI. It is a Check Point company. The platform covers AI applications, autonomous agents and employee use of external AI tools.
Legacy data loss prevention tools and web gateways cannot interpret natural language prompts or intent. Lakera targets that gap. It screens inputs and outputs for prompt injections, jailbreaks, data leaks and toxic content.
It works as a gateway layer built on the Guard API, placed between users, LLMs and external tools. Developers send prompts to the API and receive a flagged result.
Lakera also offers automated red teaming to simulate adversarial attacks before launch. Lakera reports that its detection learns from over 1 million hackers worldwide.
Pricing
Lakera offers a free Community plan at $0 per month. It includes 10,000 requests per month, an 8,000-token maximum prompt size, SaaS hosting and community support. Enterprise pricing is custom and quote-based, with flexible volumes, SaaS or self-hosted deployment, and EU or US data residency.
* Disclaimer: Please note that pricing information may not be up to date. For the most accurate and current pricing details, refer to the official website.
Key Features
- ✓
Runtime protection for AI agent actions and tool use
- ✓
Shadow AI discovery across browsers, desktop apps and IDEs
- ✓
MCP security against indirect injection through connected tools
- ✓
Detect and Enforce modes assigned through project policies
- ✓
Automated red teaming before production release
Use Cases
Protecting customer-facing chatbots
A support chatbot sends each message through the Guard API first. Injection attempts are blocked before reaching the model. Outgoing replies are filtered so toxic or off-brand text never reaches customers.
Securing RAG document assistants
An internal assistant queries HR and finance records. Lakera monitors the retrieval pipeline for leaks of sensitive data. Unauthorized access requests are blocked.
Stopping source code leaks
Employees paste proprietary code into public LLMs from browsers and IDEs. Security teams see these interactions and apply user-level policies. Context-aware data protection stops the paste.
Guarding agents connected to CRMs
An agent runs workflows in an external CRM. If an indirect prompt injection tries to trigger record deletions, Lakera blocks the action.
Testing apps before launch
A development team simulates adversarial attacks against its own application architecture. The red teaming service surfaces edge-case vulnerabilities. The team fixes them before release.
Strengths & Weaknesses
Strengths
Lakera reports runtime latency under 50 milliseconds, using chunking and parallelization for long prompts.
Lakera reports a 0.01% production false-positive rate.
It is model-agnostic, so it works across LLM providers without model changes.
Security policies apply across applications without altering application code.
It screens interactions in over 100 languages.
Weaknesses
Community caps usage at 10,000 requests per month and 8,000-token prompts.
Community excludes SSO, role-based access control and SIEM integration.
Enterprise pricing is quote-only, so budgeting requires a sales conversation.
Developers must write their own logic to handle the flagged response.
Who Is This For?
Enterprise security and IT teams: They need visibility into shadow AI use and control over what employees paste into external tools.
AI developers and MLOps engineers: They add API guardrails to chatbots and agents, and use Detect mode to observe threats first.
Compliance and risk officers: They must confirm AI applications meet data privacy, toxicity and brand safety requirements before release.
AI red teamers and QA testers: They want continuous adversarial testing to catch security drift after silent model updates.
Frequently Asked Questions
How do I connect Lakera to an existing application?
Send a POST request with the user prompt and project ID to the Guard API before calling your LLM. No changes to the model are needed.
Can I observe threats without blocking users?
Yes. Projects can run in Detect mode, which flags threats on the dashboard without interrupting the application. Switch to Enforce mode after testing.
Which threats does Lakera detect?
It covers direct and indirect prompt injections, jailbreaks, sensitive data leaks, toxic content and multilingual or multimodal attacks.
Does Lakera work for non-English applications?
Lakera supports threat detection in over 100 languages. It also accepts multimodal inputs.
How does Lakera handle employee use of public AI tools?
Workforce AI Security discovers shadow AI across browsers, desktop apps, IDEs and connected SaaS services. Administrators then apply policies at the user level.
What protection exists for autonomous AI agents?
Lakera blocks unauthorized actions, unsafe tool use and data leakage by agents. MCP security stops indirect injection before an agent acts on compromised instructions.
How do I move beyond the free Community plan?
Contact Lakera sales to upgrade. Advanced features such as SSO and RBAC come with the Enterprise plan.
What third-party recognition does Lakera cite?
Lakera was named a Representative GenAI TRiSM Vendor in a 2024 Gartner guide. The OWASP LLM and GenAI Security Landscape Guide 2025 also cites it.
Lakera provides a Guard API REST endpoint for integrating security into custom applications, agents and LLMs. The AI Gateway supports SSO, RBAC and ABAC policies through identity providers. It also connects with MCP servers to govern external tool execution. Lakera has a research partnership with Snyk on AI agent security, which is not a product integration.