AI Security Consulting
Security for teams building
with AI agents, MCP tools, and RAG
Prompt injection, data leakage, and privileged tool misuse hide in the gap between untrusted text and the tools your agents can call — where traditional security tooling doesn't look. I work with engineering and security teams to find and fix these exposures before they become incidents.
Contributor, OWASP Top 10 for Agentic Applications 2026 · CISSP · PhD · 15+ years enterprise security
Most teams' MCP integrations have never been security-reviewed for prompt injection or tool misuse. See MCP security →
AI Security Sprint — €3,500
Two working days on the one system you're least sure about: an agent, an MCP integration, or a RAG pipeline. You get an 8–10 page findings memo mapped to the OWASP Top 10 for Agentic Applications 2026. The lowest-commitment way in, and the fee credits toward a full Review.
Security Assessment
The full three-week Review: threat model, red team, and architecture review across your deployment, with an executive deck and remediation roadmap.
Learn moreFractional AI Security Officer
AI security leadership for Series B and C companies. Without the $415K hire. Async-first, three retainer tiers, monthly cadence.
Learn moreTeam Workshop
Hands-on workshops that give your engineers and security team shared vocabulary and practical skills for AI-specific risks.
Learn moreFocus areas
What I assess and secure
Specific expertise in how AI systems fail. Not generic cybersecurity rebranded.
Prompt Injection
Direct and indirect injection, jailbreaks, and multi-turn manipulation that override system intent.
Data Leakage
Paths where training data, system prompts, or user data can be extracted through model outputs.
Agentic System Risk
Tool-using agents introduce privilege escalation and confused deputy attacks. These don't exist in traditional systems.
Supply Chain & Model Risk
Third-party models, RAG data sources, fine-tuning datasets, and embedding pipelines as attack vectors.
Access Controls
How user identity, permissions, and data isolation are enforced across your AI infrastructure.
Monitoring & Response
Logging, anomaly detection, and incident response playbooks built for how AI systems actually fail.
I've spent most of my career securing critical systems in regulated industries. Power grids. Healthcare infrastructure. Financial systems.
These days I work with teams building AI agent systems. Model selection, fine-tuning, RAG pipelines, agentic workflows, production on Kubernetes. I understand how these systems are built because I've built them.
That combination puts me in an unusual position. I understand how these systems are built, and I understand how they're broken. That's the gap Molntek exists to close.
Not ready to talk yet?
Free checklists, guides, the Agent Security Scorecard, and a sample assessment report. No call required.
Ready to talk?
Tell me what you're building. I'll tell you honestly whether I can help and what that would look like.