Lakera is a company that enables enterprises to build exciting AI applications by safeguarding them from cyber risks. Pulse 2.0 interviewed Lakera CEO and co-founder David Haber to learn more about the company.
David Haber’s Background
What is Haber’s background? Haber said:
“The leadership team at Lakera has extensive experience building AI systems for high-stakes environments. My co-founder Matthias and I spent the last decade building autonomous flight systems where security and safety had to be baked into everything we did.”
“We’re now applying what we learned about deploying AI in heavily regulated industries to enterprise AI. Our team is composed of ex-Google and Meta ML engineers and combines practical expertise across AI, LLMs, and computer vision with deep regulatory and commercial experiences. I am also an adviser on the EU AI Act.”
Formation Of Lakera
How did the idea for the company come together? Haber shared:
“After developing machine learning systems for aerospace and healthcare, we saw the rapid emergence of AI models with big potential for enterprise use cases and noticed a massive white space in security. Traditional cybersecurity solutions simply don’t accommodate the unique challenges of AI. With that, our objective from day one has been to build safeguards into AI applications and systems that help enterprises safely and securely bring AI into the real world.”
Core Products
What are the company’s core products and features? Haber explained:
“Right now, our main objective is to get developers the security tools they need as fast as possible. To accommodate that urgency, this fall, we launched Lakera Guard, a security API that’s fast and seamless to deploy. In under 5 minutes, developers can integrate Lakera Guard into their AI applications and gain enterprise-grade protection.”
“The API is backed by Lakera’s threat intelligence database with tens of millions of attack data points. Our proprietary datasets grow by 100k+ entries every day, meaning Lakera Guard gets stronger over time. It’s also LLM-agnostic and currently focuses on securing Q&A use cases by protecting against prompt injections (including jailbreaks, escape characters, role play and system prompt leakage), PII detection and masking, content moderation, and hallucinations.”
“We also stress test AI applications to ensure security and safety with our red-teaming product, Lakera Red. Lakera Red specializes in detecting and identifying your AI application’s vulnerabilities. It’s designed to help organizations building with LLMs understand their attack surface prior to deployment, so that they can mitigate various risks in the application and protect the organization and their users.”
Favorite Memory
What has been your favorite memory working for the company so far? Haber reflected:
“Seeing Gandalf, our viral hacking game, become an online sensation and the go-to-resource for learning about AI security for both teams in some of the world’s biggest organizations and regular LLM users has been a really proud moment. Especially because creating Gandalf was a fun, collaborative process – born out of an internal hackathon at Lakera.”
Evolution Of Lakera’s Technology
How has the company’s technology evolved since launching? Haber noted:
“We started in the computer vision space in 2021, but soon realized that the security of LLMs was going to become a major pain point. While many businesses are excited to deploy LLM models at an enterprise scale, few understand the risks in doing so. The AI landscape is rapidly evolving and new challenges present themselves constantly. With that in mind, we’ve designed our technology to evolve in tandem. Lakera Guard, for example, is able to recognize new threats as they arise due to its always-on threat intelligence database that receives new data daily.”
Significant Milestones
What have been some of the company’s most significant milestones? Haber cited:
“Earlier this year, we raised $10 million in financing led by Redalpine. Additional investors include Fly Ventures, and Inovia Capital, and angels such as the co-founders of Snyk, the CISO at Palo Alto Networks, a SVP at Datadog, and the former CEO at Hybris. We’ve also doubled our team in the last year.”
“Since releasing our beta version of Lakera Guard in August, we have equipped thousands of developers globally to initiate the secure development of AI applications. We also recently achieved SOC 2 Type I compliance for Lakera Guard in accordance with the American Institute of Certified Public Accountants. SOC 2, which stands for Service Organization Control 2, is a security protocol and compliance standard developed by the American Institute of CPAs (AICPA). This framework ensures that service providers maintain the utmost security when handling user data. This third-party industry validation proves that we can provide enterprise-level security for our customers.”
Total Addressable Market
What total addressable market (TAM) size is the company pursuing? Haber assessed:
“We’re in novel territory sitting at the intersection of cybersecurity and artificial intelligence. Given the youth of our category, it’s tough to quantify the total addressable market size. However, the global cyber security market size was estimated at $202.72 billion in 2022 and the artificial intelligence market size was estimated at $150.2 billion in 2023.”
Differentiation From The Competition
What differentiates Lakera from its competition? Haber affirmed:
“Our biggest competitive differentiators are our proprietary AI security intelligence, developer-first delivery model, and enterprise-readiness.”
“With the AI threat landscape changing every day, enterprises need an intelligent compute layer embedded in AI applications to ensure protection over time. Lakera has curated the most advanced AI threat database in the world to power its products. We combine insights from LLM applications, our AI education game Gandalf, open-source data, and our own dedicated ML research. We’ve amassed over 30 million AI threat data points to date. And as I mentioned, this number grows every day.”
“For tasks like red-teaming AI applications, this level of insight is unparalleled. In an environment where one’s ability to thwart an attack is only possible if they are aware it exists, we need to be aware of any and all existing and developing attack methods.”
“We also design our products with developers in mind. Laker Guard fits seamlessly into developers’ current setups and enables them to instantly secure generative AI applications with one line of code in. Whether they’re using GPT-X, Claude, Bard, LLaMA or their own LLM, developers always stay in control.”
“On top of that, enterprises can deploy Lakera Guard in the cloud or on-premise, and it’s purpose-built to easily integrate across multiple use cases. This enterprise readiness is a big selling point for organizations looking to scale AI security solutions across teams. Our SOC2 compliance also gives enterprise customers reassurance that we follow the highest security and privacy standards and will ensure their data is protected at all times.”
“Product differentiation aside, our team’s AI/ML expertise also gives us a leg up. When it comes to AI security, a deep understanding of the way machine learning models work is imperative. Our background in this world has shaped both our product and approach and definitely helps set us apart.”
Future Company Goals
What are some of the company’s future company goals? Haber concluded:
“We’re ultimately working towards a future where security is intrinsic to enterprise AI. We want AI security to be as effortless as possible. Eventually, developers will need an intelligent security layer embedded in every aspect of the AI lifecycle, from development to production to maintenance and compliance. For now, we plan to evolve the product based on our insight into the AI threat landscape, and down the line we’ll be looking closer at other parts of the AI lifecycle. We are also expanding to the US with plans to open an office in San Francisco in 2024.”
