A lot of the AI security people I talk to didn't start in AI security. Most came from application security, and the move is more achievable than it looks from the outside. Here's what I tell AppSec engineers who ask me how to make it.
Why AppSec is the best starting point
Most AI security work is still application security, with a new kind of component in the middle. An LLM feature is an app that takes untrusted input, calls tools and touches data. If you already threat model web apps, review code and think in trust boundaries, you have the core of the job. What you need to add is a solid understanding of how models and agents fail.
What carries over
- Threat modeling. The single most useful skill. Every AI feature needs one, and most teams don't know how to do it well.
- Secure design and code review. Guardrails, permission checks and logging still live in code someone has to review.
- Testing like an attacker. Pentest and bug bounty instincts map straight onto prompt injection and jailbreak testing.
- Explaining risk to engineers. AI teams move fast, and people who can say "fix this first" are rare.
What you'll need to learn
- How LLM applications break: direct and indirect prompt injection, data leaking through the model, insecure output handling and agents with too much power. The OWASP Top 10 for LLM Applications is the best place to start.
- How agents change the picture: tools, permissions, memory and multi-step actions turn a chatbot bug into something that can take real actions.
- The ML pipeline basics: where training data comes from, how models are stored and served, and what supply-chain risk looks like for model files and their dependencies.
- The frameworks clients ask about: MITRE ATLAS for attack techniques, plus the NIST AI Risk Management Framework and ISO/IEC 42001 if you lean toward governance.
- Enough ML to hold the conversation. You don't need to train models. You do need to know what fine-tuning, embeddings and retrieval are, and where each can go wrong.
Build proof, not certificates
Hiring managers in this space care far more about what you've built or broken than about courses. A few weekends of real work beats a stack of certificates:
- Red team an open model, or an LLM app you built yourself, and write up what you found.
- Get hands-on with open-source testing tools such as garak and PyRIT.
- Threat model an AI feature you know well and turn it into a short write-up, leaving out anything confidential.
- Contribute to an open-source AI security project, or publish a small tool of your own.
Put the links on your CV. One good write-up does more than a paragraph of claims.
Make your CV read like AI security
Lead with the work that's closest to AI: any LLM or ML feature you secured, any agent or automation you reviewed, any testing you did on models. Use the real terms, such as prompt injection, model access controls and AI supply chain. Keep the rest of your AppSec experience, because it's still the foundation. Just don't make the hiring manager hunt for the AI part.
What the interviews look like
Expect a practical round. Common versions: threat model an LLM feature that can read customer data and call internal tools, find the problems in an agent design, walk through how you'd test a model for prompt injection, or explain a finding to an engineering team. Strong candidates think out loud, prioritize, and know when a risk is worth accepting.
Which first role to aim for
- AI security engineer at a company shipping AI features. The closest match to AppSec, and where most of the openings are.
- Product security with an AI focus. Often the easiest internal move if your current company is building AI.
- Contract red teaming or AI security assessments. A fast way to build a track record, if you're comfortable with contract work.
Research-heavy roles usually want published work or a strong ML background, so treat those as a second step rather than the first move. If you're unsure how these roles differ, our guide to AI red teamer, AI security engineer and ML security engineer roles breaks it down.
Common questions
Do I need a machine learning degree to move into AI security?
No. Most people making the move come from security, not machine learning. You need enough ML to understand how models and agents fail, which you can learn through hands-on projects and on the job.
How long does the move usually take?
It varies, but experienced AppSec engineers who put in a few months of focused, hands-on work are often credible candidates for AI security engineer roles.
Will I take a pay cut moving into AI security?
Usually not. AI security skills are scarce and priced that way. Our AI security salary benchmark has current ranges by level and region.
Thinking about the move? Send your CV to info@securaitalent.com or look through our live roles. We'll tell you honestly where you'd fit today, and your CV only goes to a client with your OK.