VFlash: Fast, Accurate Visual Planning via Flash Primitive Decoding
u_Unrolled
D19: AI Security & Safety
Phase 8: Advanced AI Domains & Specialization/DOMAIN 19

Domain 19: AI Security & Responsible AI

AdvancedUnrolled Pro

AI security and responsible AI: threat models, prompt injection and other LLM attacks, securing agents and tools, red teaming, and fairness and safety.

Estimated study time: about 60 hours
Access: Unrolled Pro
Unrolled Pro Domain

Unlock all 5 lessons with Unrolled Pro

Full production code, quantitative capacity math, failure recovery playbooks, and interview coaching.

Get Pro from ₹167/mo (~$2/mo)14-day refund guarantee

Lessons

5 lessons
01

AI Security Fundamentals — Threat Models, Poisoning, Supply Chain and Privacy Attacks

Pro 15m

Classical security still applies to AI systems (patch servers, protect secrets, authenticate users), but ML adds new assets and new attack surfaces. Training data can be poisoned so the model learns a backdoor.

✓ Runnable code✓ Exercise with solutionStart Lesson
02

LLM Attacks — Jailbreaks, Prompt Injection, Exfiltration and Red-Teaming

Pro 15m

Language models follow instructions, and they receive instructions and data through the same channel: tokens. Two families of attack follow.

✓ Runnable code✓ Exercise with solutionStart Lesson
03

Agent Security — Least Privilege, Trust Boundaries, Sandboxing and Human Confirmation

Pro 15m

Once a model can act (send, delete, pay, run code, write memory), a successful injection stops being an embarrassing answer and becomes an incident.

✓ Runnable code✓ Exercise with solutionStart Lesson
04

Responsible AI — Fairness, Privacy, Transparency and Governance in Engineering Terms

Pro 15m

"Responsible AI" sounds like a values statement, but in engineering it is a set of measurable properties and processes: error rates compared across groups, personal data handled with explicit limits, documentation that says what a model is for and where it fails, legal risk classification that decides which controls are mandatory, and humans with real authority over consequential decisions.

✓ Runnable code✓ Exercise with solutionStart Lesson
05

Bridge: From Defending the System to Advancing It

Pro 10m

Atlas is now threat-modelled, red-teamed, permission-bounded and governed.

✓ Runnable codeStart Lesson
Domain 19: AI Security & Safety · Unrolled