AI Security Skills

AI security is the part of the field where the attacker is a person rather than a statistical drift. It overlaps with conventional security and then goes somewhere conventional security has no answer for, because the vulnerability is often in the model's behavior rather than in any code you could patch.

The attack surface, in plain terms

Prompt injection. Instructions hidden in content the model reads. A support agent that summarizes a customer email can be instructed by that email. The reason this is hard is structural: the model has no reliable way to separate the instructions it was given from the data it was handed. There is no patch for that, only architecture that limits what the model is permitted to do.

Data poisoning. Corrupting training or fine-tuning data so the model learns something the attacker chose. It is slow, it is quiet, and it is very hard to detect after the fact, which makes provenance of training data a security control rather than a data quality nicety.

Model and data extraction. Querying a system enough times to reconstruct its training data or approximate the model itself. Rate limiting is not a complete answer, though its absence is a finding on its own.

Supply chain. Models downloaded from public hubs, weights of unclear provenance, and serialization formats that execute code on load. This is the closest thing to familiar territory and it is where the quickest wins usually are.

Why the traditional answers fall short

Input validation assumes you can tell good input from bad. With natural language you frequently cannot, since the malicious instruction is a legitimate sentence.

Patching assumes a fix exists. Prompt injection has no equivalent to a patched CVE, because the behavior is emergent rather than a defect in a line of code.

So the practical posture shifts toward containment. Assume the model can be made to try something, and make sure it lacks the permissions to succeed. An agent with read-only access that gets hijacked is an incident. The same agent with write access to production is a very different day.

Where to start

The OWASP Top 10 for Large Language Model Applications is the most usable reference for practitioners, and MITRE ATLAS catalogues adversarial techniques against machine learning systems in a structure security teams already recognize from ATT&CK. NIST has published adversarial machine learning taxonomy work that is useful for shared vocabulary.

Naming these accurately in an interview is a reasonable proxy for having done the reading, and most candidates cannot.

The governance side pays too

Not everyone in this area is doing red team work. A large share of the hiring is for people who can assess whether an AI deployment is sensibly bounded: what the system is allowed to reach, what happens when it is wrong, who reviews its actions, and whether anyone would notice an abuse.

Those are governance questions with a security frame, and people from GRC backgrounds answer them better than people who only know the attacks.

Practising without a lab

Work out the blast radius of an AI feature you already use. What can it read, what can it change, who would see it if it went wrong. Write that up in a page. It is a more convincing artifact than a certificate, and it is the exact analysis the job consists of.

Where to go next

• Browse the jobs that use these skills
• Follow a career roadmap into the role you want
• Hiring for this? Start from a job description template
• Free certification study games, 592 practice questions

More in this series

• 9 Essential Data Governance Skills for the AI Era
• 10 Internal Audit Skills for Modern Assurance Careers
• 12 Transferable GRC Skills You May Already Have
• Technical vs. Nontechnical GRC Skills: What Employers Actually Need
• AI Governance Skills Employers Actually Hire For
• GRC Analyst Skills: What the Job Actually Requires
• Compliance Analyst Skills
• Risk Assessment Skills
• Controls Testing Skills
• Policy Writing Skills
• Regulatory Change Management Skills
• Third-Party Risk Skills
• Model Risk Management Skills
• AI Impact Assessment Skills
• AI Auditing Skills
• AI Evaluation and Testing Skills for Governance Careers
• Data Lineage Skills
• Data Quality Skills
• Privacy Engineering Skills
• AI Incident Response Skills
• Governance Program Management Skills
• Stakeholder Communication Skills
• Executive Risk Reporting Skills
• Evidence Documentation Skills
• Control Mapping Skills
• Framework Crosswalking Skills
• Vendor Due Diligence Skills
• Responsible AI Skills
• GRC Tools and Automation Skills
• How to Build the 9 Data Governance Skills: A 12-Month Career Plan