Working remotely with SaaS teams across DACH, the UK, and the US.
Sebastian Tagwercher
AI & LLM Security Audits, grounded in research.
I audit the security of LLM-powered features — chatbots, copilots, RAG pipelines, agentic systems — backed by published MSc research into attack vectors against large language models. Web pentesting also available via my vetted partner network.
My MSc thesis on LLM attack vectors developed a formal taxonomy of attacks against, using, and within large language models. It combines a comprehensive literature review with four practitioner interviews and an empirical exploration of a Cambridge University dark-web forum dataset. The taxonomy now shapes how I scope and structure security audits.
Read the research →Article 15 of the EU AI Act, now deferred to December 2027 for standalone high-risk AI systems, requires documented resilience against the same attack categories the thesis taxonomy describes.
Services
One primary practice. Web pentesting available as a direct referral.
From €15,000
Scoping calls from €2,500
3–6 weeks depending on scope
- ✓OWASP LLM Top 10 2025 + Agentic 2026 coverage
- ✓Prompt injection, RAG, and agent tool-call testing
- ✓CVSS-rated findings mapped to NIST AI RMF and EU AI Act
Via referral
Need a standalone web pentest? I'll refer you to senior testers I trust — no margin, direct introduction.
Why work with me
There are plenty of security consultants. Here's what makes this different.
Published research on LLM security
Not a certification badge — peer-reviewed MSc work analyzing real attacker behavior in dark-web forums, with a formal taxonomy of attacks against, using, and within LLMs. Read the research.
Business fluency
BA in Business Administration and corporate tax accounting background. Findings connect to your P&L and your risk register, not just your tech stack.
Research-backed methodology
Six-phase methodology grounded in OWASP LLM Top 10 2025, OWASP Agentic 2026, MITRE ATLAS, and NIST AI RMF Generative AI Profile. Tooling: Garak, Promptfoo, PyRIT, DeepTeam, plus manual testing.
Fixed scope, honest pricing
Every engagement is scoped upfront with a written SOW and a fixed price band. 50/50 billing. No hourly overruns, no scope creep invoices.
See what a real report looks like
Before you commit, see exactly what you'll get. The sample report shows the format, severity ratings, reproduction steps, and remediation guidance from a real AI security engagement.
Case study
What a full AI security engagement looks like
From first contact through scoping, testing, and retainer — a sanitized composite of a real engagement, showing the full methodology and timeline.
Get in touch
I'll respond within 24 hours. If you want to skip the form — email me directly or book a 20-min intro call.