Agentic AI security
Agentic AI security and security automation
AI agents read your data, call your tools and act on your behalf. I test them the way an attacker would and put guardrails around what they can do, then put the same technology to work on security itself.
When to call
When teams ask for it.
- An AI agent is about to get access to production data, email or internal tools
- Teams are adopting AI tools faster than security can review them
- Analysts spend their day on work a workflow could do
Scope
What it involves.
- Threat modelling AI agents: what they can read, which tools they can call and who can steer them
- Testing agents and LLM features for prompt injection, data leakage, excessive agency and unsafe tool use
- Guardrails: least-privilege tool access, logging and human approval for actions that matter
- Designing automated workflows (n8n and similar) for repetitive security tasks
- MLSecOps: security across the model lifecycle, from training data to monitoring
How it works
How the work is done.
Untrusted words in, real actions out.
I trace each route from text an attacker controls to a tool your agent can call, then try the ones that matter.
- Normal request
- Injected instruction
- Policy guard
Outcome
What you have at the end.
AI agents and automation you can explain to an auditor: tested, limited to what they need, and logged.
Usually for: Teams building or adopting AI agents and LLM features, or automating operations
Proof
Work and credentials behind it.
Other services
Other ways I can help.
- Penetration testingWeb, API, mobile, network and cloud testing from an attacker's point of view.
- Security operationsDetection engineering, SIEM, incident handling and malware analysis.
- GRC advisoryInformation system audits, ISO/IEC 27001 and security strategy.
- Training and speakingSecurity awareness, technical workshops and conference talks.
Let's scope it together.
Tell me what you need to protect and by when. I'll reply with whether I can help and how I'd approach it.