AI Security Engineer
A specialized programme focusing on securing AI-powered applications, understanding vulnerabilities in modern AI systems and developing practical AI security skills.
Overview
Applications built on large language models fail in ways traditional security testing does not look for. The input is language, the boundary between instruction and data is blurred, and the model is not deterministic.
This programme covers how these systems are put together, where they break, and how to test them properly: prompt injection, jailbreaks, retrieval poisoning, insecure tool use and the API surface around a model.
The work is applied. The aim is to leave able to assess a real AI application and write up what you found.
At a glance
- Focus
- AI Security
- Modules
- 12 proposed modules
- Practical work
- 5 lab exercises, 3 projects
- Prior experience
- No machine learning background assumed
Duration, fees and any form of certification are not published here because they are not yet established. Ask us directly and we will tell you exactly where things stand.
Skills you develop
- Explain how an LLM application is assembled and where trust boundaries sit
- Construct and test prompt injection, both direct and indirect
- Assess retrieval pipelines for poisoning and leakage
- Review the API and tool-calling surface around a model
- Run a structured red team exercise against an AI application
- Apply the OWASP Top 10 for LLM Applications to a real system
- Threat model an AI feature before it ships
Curriculum
Proposed programme content. Expand a module to see what it covers.
Tools and technologies
Hands-on labs
- Extract a system prompt from a deployed chat application
- Plant an indirect injection in a document a RAG pipeline will retrieve
- Abuse a tool-calling agent into an action outside its intended scope
- Find and exploit a weak authorisation boundary on an AI API
- Poison a retrieval corpus and measure the effect on answers
Projects
Run a full red team assessment of a vulnerable AI application and report the findings
Threat model an AI feature and propose proportionate controls
Build a small test harness that checks an LLM endpoint against a set of known attack patterns
What you can do by the end
- Assess an AI application and explain how it fails, not just that it does
- Produce a reproducible red team write-up
- Judge which mitigations are real controls and which only look like one
- Threat model an AI feature at design time
Roles this prepares you for
These are the kinds of roles the programme is aimed at. PolarSight does not guarantee employment, placement or an interview.
Prerequisites
- Working Python, enough to read and modify a script
- Basic web and API understanding, including HTTP and authentication
- No machine learning background assumed
Who this is for
- Security practitioners moving into AI security
- Developers building on language models who need to secure what they ship
- Students with a programming foundation seeking a specialised route
Want to run AI Security Engineer with your batch?
Send us the details and we will reply with a programme outline and what delivery would involve.
SOC Analyst
A practical cybersecurity training programme designed to develop skills in security monitoring, threat detection, security incident investigation and incident response.
View programme →