Insights
Analysis of AI-security threats, autonomous-agent risk, and the techniques attackers actually use, published through KevinBytes.
Latest / Agentic AI
AI-agent risk is not just prompt injection. Once an agent can call tools, touch data, modify workflows, or influence CI/CD, it becomes part of the production...
Read latestField guide
A visual guide to the 2025 OWASP categories for AI agents, RAG, MCP, and tool-connected applications.
Open the OWASP guidePractical resources
Archive
Showing 12 of 12 posts.
AI-agent risk is not just prompt injection. Once an agent can call tools, touch data, modify workflows, or influence CI/CD, it becomes part of the production...
A pre-launch review list for AI agents that touch production APIs, customer data, tools, memory, or RAG context.
A security checklist for MCP servers, clients, OAuth flows, tokens, tools, permissions, trust boundaries, logging, and blast radius.
AI agent blast radius is the maximum plausible damage an agent can cause if manipulated, misconfigured, over-permissioned, or exposed to hostile context.
How to review RAG systems for authorization failures, tenant-isolation gaps, prompt injection, vector-store leakage, document poisoning, and audit logging.
How founders, CTOs, CISOs, and product-security teams can choose between an AI red team, LLM pentest, AI security assessment, or production readiness review.
How engineering and product teams can prepare AI agents for enterprise security review, production launch, customer diligence, and governance scrutiny.
An MCP threat model for teams connecting LLMs and AI agents to tools, OAuth tokens, APIs, local servers, SaaS systems, and production workflows.
How indirect prompt injection reaches AI agents through RAG systems, copilots, MCP tools, webpages, emails, PDFs, memory, and tool outputs.
A framework for classifying AI agent tool permissions by business impact, authorization boundary, required controls, and production readiness.
Common RAG authorization failures, tenant-isolation gaps, vector-store access-control mistakes, source attribution issues, and safer retrieval design.
How disciplined engineering assessment helps legal, insurance, and executive teams understand root cause, technical exposure, and the next defensible decision...
