The Evolution of AI Risk

Explore how AI risk has evolved from traditional web vulnerabilities to the unique architectural challenges of Large Language Models. This video traces the journey from early training data memorization to the modern-day risks of "vibe coding" and autonomous agent exploitation
.
We break down the fundamental shift in the threat landscape, including:
The Architecture Gap: Why the Transformer’s "attention mechanism" creates a permanent lack of trust boundaries between system prompts and user input
.
The Rise of LLM-Specific Risks: How the OWASP Top 10 for LLMs replaced traditional security models to address threats like prompt injection and training data poisoning
.
From Bots to Agents: The evolution of risk as AI moves from conversational tools to autonomous actors using the Model Context Protocol (MCP) and multi-step orchestration
.
Vibe Coding: Why relying on "correct-looking" AI-generated code is the new frontier for inherited software vulnerabilities
.
The Future of Defense: Implementing multi-layered "Defense-in-Depth," from perplexity filtering to human-in-the-loop (HITL) gates
.
Whether you are a security engineer, a red teamer, or a developer, understanding this evolution is critical for building secure AI-powered products
.
Key Topics Covered: ✅ Transformer Architecture & Security Relevance
✅ The "Alignment Tax" and Jailbreak Taxonomy
✅ Indirect Prompt Injection & RAG Security
✅ Governance Frameworks: NIST AI RMF & the EU AI Act

Видео The Evolution of AI Risk канала Juddy_TalksCybergrc