Lakera
LLM security platform providing prompt-injection detection and real-time guardrails for generative AI applications, acquired by Check Point.
Visit Website ↗Overview
Lakera builds real-time detection and guardrail infrastructure for large language model applications, focused on prompt injection, jailbreaks, data leakage and unsafe outputs. Its Lakera Guard product sits between an application and its LLM as an inline filter, and the company is also known for Gandalf, a widely played public game that crowdsources adversarial prompts and doubles as a research and marketing tool for prompt-injection techniques.
Founded in Zurich in 2021 by a team with backgrounds at Google and Meta AI, Lakera raised roughly $30M through a Series A before being acquired by Check Point Software in September 2025 for a reported deal value around $300M, with Lakera’s Zurich site becoming Check Point’s AI security R&D hub. The acquisition folds Lakera’s detection engine into a much larger, established network-security vendor’s portfolio rather than leaving it as a standalone company.
Innovation Matrix Assessment
Shipped a widely adopted inline guardrail product and used the public Gandalf game to continuously harvest new attack patterns, in a category that didn't exist three years ago.
Guard is deployed as a production filter in front of live LLM apps, giving security teams a concrete control point rather than a purely theoretical framework.
Acquired by Check Point for a reported ~$300M in September 2025, a strong independent signal of market validation, following a ~$30M Series A.
Addresses prompt injection and LLM-specific abuse patterns that have no direct analog in traditional AppSec tooling.
No independently published third-party red-team benchmark of Guard's detection accuracy was found; Gandalf demonstrates attack collection, not defense efficacy.
Prompt injection is recognized as one of the top LLM application risks (OWASP LLM Top 10), so the problem space is durable as GenAI adoption grows.
Why CISOs Should Care
Gives security teams an inline control point to block prompt injection and data leakage in production LLM applications rather than relying only on model-provider defaults.
What Makes It Different
Uses a large, continuously growing public adversarial-prompt dataset (from the Gandalf game) to train detection models, instead of a static internal red-team corpus.
The Matrix Verdict
72/100 — MEANINGFUL INNOVATOR
A Strong Performer within this young category: real acquisition-level validation and a genuinely AI-native product, but independent efficacy evidence is thin, which caps the overall score below Transformational tier.
Editorial Note: Claims vs. Verified Findings
The $300M acquisition price and Zurich R&D-hub status are reported by trade press, not confirmed in an official financial disclosure. Detection-accuracy claims on Lakera's own site are UNVERIFIED vendor marketing; no independent efficacy study was found.
Sources
Alternatives to Lakera
Palo Alto Networks (Prisma AIRS)
Major cybersecurity incumbent's unified AI security platform (Prisma AIRS) for discovering, assessing and protecting AI models, agents and…
Zenity
Governance and security platform for AI agents and low-code/no-code development, securing agent identity, permissions and behavior across the…
Cisco AI Defense (Robust Intelligence)
AI security product line built on Robust Intelligence's algorithmic red-teaming and AI Firewall technology, now integrated into Cisco's…
HiddenLayer
AI security platform unifying ML supply-chain scanning, runtime defense, posture management and automated red-teaming for enterprise AI systems.
Protect AI (now part of Palo Alto Networks)
AI/ML security company (model scanning, MLSecOps, AI-SPM) acquired by Palo Alto Networks for $634.5M and folded into Prisma…
Microsoft (AI Security portfolio)
Major cloud/software incumbent extending its Purview, Defender and Security Copilot products to cover AI-specific data-security, posture and threat-detection…