Amit Sides (@amitsides)

A Multidisciplinary Theory of Exploiting LLMs and Guardrails: The Collapse of Embedding Geometry

Sep 21, 2025 · 13 min readai-blog.dev

Agentic AI SecOps Overview Part 1: Between Red-Blue Multi-Agents ContextSecOps Bottle and Poison

I will now present some AI SecOps use cases. The first one is how to build two MAS (Multi-Agent Systems) as Red/Blue teams, where there are multiple roles. The Red team can be used for prompt stress testing (Prompt Metaspolit if you’d like), and the ...

Aug 25, 2025 · 11 min readai-blog.dev

AI Agents and their Ontologies, Psychologies, RAG Capabilities, Identity, Personality, Communication Skills, and Infrastructure

*You’re under no obligation to be the same person you were five minutes ago. (*Alan Watts) Overview: From Prompts to Agentic Identity I have seen different prompts instructing Agents to play a role. Many agents with different roles converse with eac...

Mar 29, 2025 · 10 min readai-blog.dev

Cyber Unicorns Reverse Engineered (CSPM & XDR): JSON, Cypher, Graphs, Embeddings, and Intruders

Tags: AI, Cyber, CSPM, XDR, GraphMap, GNN, Cypher, XGBoost, Python What’s the technology behind Cybersecurity Unicorns? What are the complexities of building AI First XDR/CSPM Products? Multiple Use Cases, Misconfigurations, Datatypes, and Cyber Ont...

Mar 18, 2025 · 8 min readai-blog.dev

RAG Agnostic Agentic AI: The OmniScience Design Pattern

We all want our AI Agents to be database agnostic, query languages agnostic, and API-calling Agnostic. With the help of LLM, this is now possible. We can, as LLMs, build queries on the fly for our RAG to retrieve relevant data and then send it to LLM...

Feb 17, 2025 · 8 min readai-blog.dev

Amit Sides

About

Available for

Amit Sides's blogs

A Multidisciplinary Theory of Exploiting LLMs and Guardrails: The Collapse of Embedding Geometry

Agentic AI SecOps Overview Part 1: Between Red-Blue Multi-Agents ContextSecOps Bottle and Poison

AI Agents and their Ontologies, Psychologies, RAG Capabilities, Identity, Personality, Communication Skills, and Infrastructure

Cyber Unicorns Reverse Engineered (CSPM & XDR): JSON, Cypher, Graphs, Embeddings, and Intruders

RAG Agnostic Agentic AI: The OmniScience Design Pattern

Amit Sides

About

Available for

Amit Sides's blogs

A Multidisciplinary Theory of Exploiting LLMs and Guardrails: The Collapse of Embedding Geometry

Agentic AI SecOps Overview Part 1: Between Red-Blue Multi-Agents ContextSecOps Bottle and Poison

AI Agents and their Ontologies, Psychologies, RAG Capabilities, Identity, Personality, Communication Skills, and Infrastructure

Cyber Unicorns Reverse Engineered (CSPM & XDR): JSON, Cypher, Graphs, Embeddings, and Intruders

RAG Agnostic Agentic AI: The OmniScience Design Pattern

Amit Sides

About

Available for

Amit Sides's blogs

A Multidisciplinary Theory of Exploiting LLMs and Guardrails: The Collapse of Embedding Geometry

Agentic AI SecOps Overview Part 1: Between Red-Blue Multi-Agents ContextSecOps Bottle and Poison

AI Agents and their Ontologies, Psychologies, RAG Capabilities, Identity, Personality, Communication Skills, and Infrastructure

Cyber Unicorns Reverse Engineered (CSPM & XDR): JSON, Cypher, Graphs, Embeddings, and Intruders

RAG Agnostic Agentic AI: The OmniScience Design Pattern

Search Hashnode

Amit Sides

About

Available for

Amit Sides's blogs

A Multidisciplinary Theory of Exploiting LLMs and Guardrails: The Collapse of Embedding Geometry

Agentic AI SecOps Overview Part 1: Between Red-Blue Multi-Agents ContextSecOps Bottle and Poison

AI Agents and their Ontologies, Psychologies, RAG Capabilities, Identity, Personality, Communication Skills, and Infrastructure

Cyber Unicorns Reverse Engineered (CSPM & XDR): JSON, Cypher, Graphs, Embeddings, and Intruders

RAG Agnostic Agentic AI: The OmniScience Design Pattern