Manifold

Neal built the most widely adopted open-source LLM firewall in the world, watched it get acquired into Palo Alto Networks, and then concluded that everything he'd built was already insufficient for the next problem. That is the founder trajectory that produces genuinely category-defining companies: someone who solved the previous version of a problem at scale, saw the limitation of that solution, and built the next one. LLM Guard, the product he co-built at Laiyer AI, was downloaded 2.5 million times per month on HuggingFace before the acquisition. That is not a research project that got traction. That is production infrastructure that became the default choice for teams securing LLM applications. Neal understands what it takes for a security tool to achieve that kind of adoption because he did it once already. His intellectual shift from Laiyer to Manifold is specific and worth understanding before you contact him. LLM Guard secured the inference perimeter: it monitored what went into the model and what came out. That architecture made sense when AI systems were chatbots, tools that generated text and then stopped. It does not make sense for agents that execute shell commands, read codebases, make API calls, and take real-world actions after the inference point. Neal's argument is that the security industry has responded to the rise of agents by scaling the chatbot-era tools, adding more classifiers, more guardrails, more gateways, and that this approach is fundamentally blind to the actual risk surface. Manifold's bet is that you need a new category of tool designed specifically for the agent-action layer, not a retrofitted version of what worked for text generation. That is a precise and defensible thesis, not a general "AI security is growing" pitch. He is a European founder, educated at HEC Paris, based in the Berlin metropolitan area based on his LinkedIn profile, who has moved to San Diego to build Manifold with his co-founders. That geographic decision, choosing San Diego over San Francisco, signals a team that is building around specific people rather than chasing the Bay Area talent pool. His communication style, based on the public quotes available from the funding announcement, is direct and specific: he speaks in concrete technical problems, not in market size claims. "Every developer today has coding agents on their laptop with access to source code, production systems, and CI/CD pipelines connected to an expanding ecosystem of MCP servers, skills, and third-party tools that no one is inspecting" is a grounded, specific threat description, not a category headline. Match that register when you reach out.

Oleksandr co-built LLM Guard with Neal and is the engineering architect whose work made the most widely deployed open-source LLM firewall in the world actually run at production scale. He has been shipping AI security code in public, the LLM Guard GitHub repository is his most detailed public profile, which means you can read exactly how he thinks about security problems technically before reaching out. His work on LLM Guard covers input/output sanitisation, prompt injection detection, PII redaction, and real-time observability tooling, all built with the constraint that it had to work reliably in production environments across every major LLM provider. That background maps directly to the signal processing and detection architecture work at the heart of Manifold. His public LinkedIn profile is sparse by design. GitHub activity and the Laiyer/Protect AI codebase history are your best window into his thinking. For an engineering hire, he is the peer-level evaluator of your technical proof of work. A security research document or a minimal endpoint monitoring prototype that demonstrates you understand the detection problem at the agent-action layer will reach him far more effectively than a resume.

Michael met Neal and Oleksandr after the Laiyer AI acquisition, when all three were working inside Protect AI and watching the same pattern develop: the AI agent use case was expanding faster than any existing security tooling could accommodate, and the response from incumbent vendors was to retrofit chatbot-era tools for a fundamentally different problem. His specific role at Protect AI and before is less publicly documented than his co-founders', but his involvement from the inception of Manifold, not as a later addition, but as a co-founder at the point of the original thesis development, signals that he brings something the other two don't have: likely enterprise GTM and security customer-facing experience that balances Neal's product background and Oleksandr's engineering depth. For enterprise security GTM and solutions engineering roles, Michael is the most relevant first contact after Neal. Find him on LinkedIn by searching his name alongside Manifold Security.