Your First 90 Days
in AI Security
A realistic week-by-week plan — not a fantasy sprint. This timeline has 20% overhead built into every phase. Life happens. You’ll miss a day. A tool won’t install cleanly. A concept will take longer than expected. That’s not failure — that’s how learning actually works.
The 20% Overhead Rule
Most 90-day plans assume you’ll execute perfectly every single day. That’s not how career transitions work. You have a current job, personal obligations, bad days, and the inevitable frustration of getting a new tool to compile on your machine.
This plan schedules ~72 days of actual work across 90 calendar days. The remaining 18 days are distributed as buffer throughout — not lumped at the end. Every phase includes explicit catch-up time. If you finish a week early, use the buffer for deeper exploration. If you fall behind, you’re still on track.
Phase 1 (Days 1–30): 24 days of planned work + 6 buffer days. If week 3 bleeds into week 4, that’s by design.
Phase 2 (Days 31–60): 24 days of planned work + 6 buffer days. Certification study may take longer than guides suggest.
Phase 3 (Days 61–90): 24 days of planned work + 6 buffer days. Portfolio building is iterative, not linear.
The market data is on your side. ISC2 projects 4.8 million unfilled cybersecurity positions globally. AI security is the fastest-growing subset. The BLS projects 29% growth for infosec analysts through 2034. You don’t need to be perfect in 90 days. You need to be credible, competent, and moving.
Sources: ISC2 Workforce Study 2025 (4.8M unfilled); BLS Occupational Outlook Handbook (29% growth 2024–2034)Before Day 1 — Prep Checklist
Don’t start the clock until these are in place. This prep work prevents wasted days in Phase 1 debugging environment issues instead of learning AI security.
Foundation: Days 1–30
Tool installation issues, Python dependency conflicts, and “what does this error mean” moments are expected, not failures. The buffer covers 2–3 days of environment troubleshooting and 3–4 days of concept review when something doesn’t click on first pass.
OWASP LLM Top 10 — Read, Understand, Map
- Read the full OWASP LLM Top 10 (2025) — all 10 items, not just summaries. Spend 1–2 hours per item.
- For each risk, write one paragraph in your journal: What it is, how it works, how you’d detect it, and which role owns mitigation.
- Install Garak (NVIDIA’s LLM vulnerability scanner). Run it against a local model or API endpoint. Document what you find.
- Cross-reference: Read our Frameworks & Practices Deep Dive OWASP section for role mapping.
MITRE ATLAS — The Adversarial Playbook
- Study MITRE ATLAS: 14 tactic categories in an ATT&CK-style matrix. Focus on understanding the attack chain, not memorizing technique IDs. Tactic IDs use the AML.TA#### format; technique IDs use AML.T####.
- Read documented case studies on atlas.mitre.org. For each: What was the target? What tactic was used? What would have prevented it?
- Install PyRIT (Microsoft’s Python Risk Identification Toolkit). Run a basic red team test against an LLM endpoint.
- Map ATLAS tactics to OWASP risks. Where do they overlap? Where does ATLAS go deeper?
Hands-On: Your First Vulnerability Assessment
- Install ART (Adversarial Robustness Toolbox). You now have three tools: Garak, PyRIT, ART. Run each against a test target.
- Write your first vulnerability report. Pick one finding from your tool runs and document it: vulnerability, impact, evidence, remediation.
- Register for the next AI Village CTF (Capture The Flag) competition, or start a HackTheBox AI challenge if available.
- Begin community engagement: Join relevant Discord/Slack channels. Lurk first. Contribute when you have something to add.
Consolidation & Catch-Up
- Review your learning journal. What’s solid? What still feels shaky? Spend 2–3 days revisiting weak areas.
- If on track: Begin exploring governance frameworks — skim NIST AI RMF overview (free at nist.gov). This previews Phase 2.
- If behind: Use this entire week as buffer. Finish the ATLAS case studies. Get all three tools running cleanly. No guilt.
- Update your LinkedIn with AI security keywords. Follow thought leaders. Signal your transition to your network.
By day 30, you should be able to: explain all 10 OWASP LLM risks from memory (not perfectly — from understanding), describe the ATLAS attack chain at a high level, run at least one AI security tool against a test target, and have a learning journal with 20+ entries. If you hit 3 out of 4, you’re ready for Phase 2.
Acceleration: Days 31–60
Certification study guides consistently underestimate real study time by 20–30%. If a cert says “40 hours,” plan for 50–52 hours. The buffer days absorb this. Don’t sacrifice depth for speed.
Governance Frameworks & Certification Selection
- Study NIST AI RMF 1.0 (NIST AI 100-1, January 2023) in depth. Understand the four core functions: GOVERN (organizational risk culture), MAP (context & categorization), MEASURE (TEVV & metrics), MANAGE (risk treatment & response). Also learn the 7 trustworthiness characteristics. This is the backbone of organizational AI risk.
- Decide on your first certification based on your background and target role:
• Offensive track: CAISP ($999–$1,099 lifetime, as of April 2026) or HackTheBox AI Red Teamer ($490/yr)
• Governance track: AIGP ($649–$799) by IAPP
• Foundation track: CompTIA Security+ ($404–$425) if no security baseline exists - Enroll and begin structured study. Block 1–2 hours daily on your calendar.
Deep Certification Study + Practical Application
- Continue certification study (15–20 hours this week). Complete first major module or domain.
- Apply what you’re learning: If studying CAISP, run labs from the 30+ hands-on exercises. If studying AIGP, draft a mock AI governance policy.
- Read the EU AI Act security provisions if targeting compliance roles. Understand risk classification tiers and obligations for high-risk AI systems.
- Participate in your first CTF challenge or complete a HackTheBox AI security challenge. Document your approach and findings.
Portfolio Building Begins
- Create a GitHub repository for your AI security work. Structure it: /tools (scripts you’ve written), /reports (vuln assessments), /notes (framework summaries).
- Write up 2–3 vulnerability assessments from your tool runs into professional-format reports.
- Continue certification study (15–20 hours). You should be 50–60% through the material.
- If applicable: Register for bug bounty platforms (HackerOne, Bugcrowd). AI-specific bounties are growing. OpenAI offers bounties up to $100K.
Consolidation & Certification Push
- Final certification study push. Complete remaining material. Begin practice exams or lab reviews.
- If using buffer time: Review all Phase 1 material. The OWASP and ATLAS knowledge should feel natural, not memorized.
- Network checkpoint: Have you connected with 5+ people in AI security? Attend a virtual meetup or webinar this week.
- Review your learning journal. You should see clear progress from Day 1. Highlight the transformation for future interview storytelling.
By day 60, you should have: certification study 70%+ complete (or exam scheduled), 2–3 professional vulnerability reports in a GitHub portfolio, at least one CTF attempt or bug bounty submission, and working knowledge of at least one governance framework (NIST AI RMF, ISO 42001, or EU AI Act). Hit 3 out of 4 and you’re on track.
Specialization: Days 61–90
Job search prep takes real time. Resume rewrites, LinkedIn optimization, cover letter drafts, and informational interviews are work, not afterthoughts. The buffer accounts for this. Don’t treat job prep as “extra” — it’s core Phase 3 work.
Certification Completion & Role-Specific Depth
- Take the certification exam, or finalize study for a scheduled date. If CAISP: complete the 6-hour practical + 24-hour report.
- Begin role-specific specialization. Pick your target role and go deep on the skills specific to it:
• AI Red Teamer: Advanced ATLAS techniques, CTF rankings, adversarial example generation
• AI Security Engineer: Pipeline security, model validation, OWASP defense implementation
• AI GRC Analyst: ISO 42001 audit prep, compliance mapping, policy documentation - Read 3–5 recent AI security incident reports. Analyze each with your framework knowledge.
Portfolio Polish & Community Contribution
- Finalize your GitHub portfolio. Add README files that explain your methodology, not just results.
- Write a blog post or LinkedIn article about something you learned. “How I Used Garak to Find Prompt Injection Vulnerabilities” or “MITRE ATLAS: What Traditional Security Misses About AI Threats.”
- Make your first meaningful community contribution: answer a question in a forum, submit a tool improvement, or share a writeup from a CTF.
- If targeting specific companies, research their AI security programs. Tailor your portfolio to demonstrate relevant skills.
Job Search Launch
- Rewrite your resume with AI security focus. Lead with skills and certifications, not just job titles. Quantify impact where possible.
- Optimize LinkedIn: headline should include “AI Security” + your specific angle. Featured section should link to your GitHub and published content.
- Begin targeted applications. Focus on roles that match your 90-day specialization:
• Offensive track: AI Red Teamer — CTF portfolio and ATLAS fluency are critical differentiators. Adversarial ML Researcher roles command $157K–$222K.
• Governance track: AI Model Risk Analyst: $100K–$160K (SR 11-7 + AI expertise in banking sector)
• Infrastructure track: AI Infrastructure Security Specialist: $160K–$240K (OpenAI, NVIDIA, CoreWeave hiring) - Schedule 2–3 informational interviews with people in AI security roles. Ask about their first 90 days.
Consolidation & Forward Planning
- Review your entire 90-day journey. Update your learning journal with a summary: where you started, what you learned, what’s next.
- If certification exam is still pending: finalize prep and schedule it for the next 2 weeks. You’re ready.
- Plan the next 90 days. Your first 90 built the foundation. The next 90 is about: advanced certifications, deeper specialization, or landing the role.
- Continue job applications and networking. The pipeline you built in Week 11 takes 4–8 weeks to generate interviews. Keep feeding it.
By day 90, you should have: at least one AI security certification earned or exam scheduled, a public GitHub portfolio with 3+ professional artifacts, one published piece of content (blog, LinkedIn article, or CTF writeup), and active job applications submitted. If your certification exam is scheduled for day 100 instead of day 85, you’re still on track. The 20% overhead means your “day 90” might land on calendar day 108. That’s the plan working, not the plan failing.
You will not be a senior AI security engineer on day 90. You will be a credible, demonstrably competent professional who can speak the language, use the tools, understand the frameworks, and show evidence of applied learning. That’s what gets you hired. The market has 4.8 million unfilled cybersecurity positions. Only 14% of organizations report having adequate AI security talent. Companies aren’t waiting for perfection — they’re looking for people who can learn fast and contribute quickly. Your 90-day portfolio proves you can do both.
Sources: ISC2 Workforce Study 2025 (4.8M unfilled globally); World Economic Forum 2025 (14% adequate AI security talent)Essential Tools & Environment
These are the tools you’ll install and use across the 90 days. All are open source or have free tiers. Total cost for the tool stack: $0.
Quick Start by Background
The 90-day timeline above is the general plan. Here’s how to adapt it based on where you’re starting. Each path adjusts the emphasis — the timeline stays the same, but what you prioritize in each phase shifts.
The background that adapts fastest to AI security isn’t always the most technical one. GRC professionals who learn enough technical AI to ask the right questions are in massive demand — because organizations with $30.9 billion in AI security spending need people who can connect technical controls to business risk and regulatory obligations. Don’t undersell your non-technical expertise.
AI-in-cybersecurity market: $30.9B (2025), 22–24% CAGR • Sources: StationX, Cybersecurity Ventures 2025