Lakera is an AI security company that specializes in protecting large language model applications from prompt injection attacks, jailbreaking attempts, data leakage, and other LLM-specific security threats. Founded in 2021 by David Haber, Matthias Kraft, and Severin Elvatun, and headquartered in Zurich, Switzerland, Lakera focuses specifically on the security challenges unique to generative AI applications. The company's flagship product, Lakera Guard, provides a real-time API that sits between user inputs and LLM applications to detect and block malicious prompts before they reach the model. Lakera Guard analyzes incoming prompts for various attack patterns including direct and indirect prompt injection, jailbreak attempts, harmful content requests, and attempts to extract sensitive information or system prompts from the model. The system uses proprietary machine learning models trained on a continuously updated dataset of attack techniques, including data from Gandalf, Lakera's public prompt injection game that has collected millions of adversarial prompt examples from users worldwide. Lakera Guard also monitors LLM outputs for sensitive data leakage, toxic content, and off-topic responses, providing a comprehensive input-output security layer. The API is designed for low-latency integration, adding minimal overhead to LLM application response times. Lakera provides integrations with major LLM providers and frameworks, and can be deployed as a cloud API or on-premises for organizations with strict data residency requirements. The platform includes a dashboard for monitoring security events, analyzing attack patterns, and configuring detection policies. Lakera offers a free tier for developers with limited API calls, a paid tier for production applications, and custom enterprise pricing for organizations requiring higher volumes, dedicated support, and on-premises deployment.

AI Content Moderation

Lakera Guard monitors both inputs to and outputs from LLM applications, detecting and filtering toxic content, harmful requests, and policy-violating responses. This input-output moderation layer helps organizations maintain content safety standards in their AI applications, preventing both intentional misuse and unintended harmful outputs.

AI Cybersecurity

Lakera provides AI-specific cybersecurity through Lakera Guard, which protects LLM applications from prompt injection attacks, jailbreaking, and data extraction attempts. Its real-time API inspects prompts before they reach the model and monitors outputs for sensitive data leakage, providing a security layer purpose-built for the unique threat landscape of generative AI applications.

AI Safety Tools

Lakera contributes to AI safety by preventing misuse of LLM applications through real-time detection of harmful prompts, jailbreak attempts, and policy-violating inputs. Its proprietary models, trained on millions of adversarial examples from its Gandalf prompt injection game, help organizations ensure their AI applications cannot be manipulated into producing harmful outputs.

AI Testing Tools

Lakera helps organizations test their LLM applications for security vulnerabilities through adversarial testing methodologies informed by millions of real-world attack examples. Its platform enables security teams to evaluate how their AI applications respond to prompt injection, jailbreaking, and other attack vectors before deployment.