Tags

• series 11
• llm 2
• tensors 2
• mechanistic-interpretability 2
• red-teaming 2
• embeddings 2
• neural-networks 1
• mathematics 1
• deep-learning 1
• attention 1
• transformers 1
• adversarial-ml 1
• attack-surface 1
• model-security 1
• reverse-engineering 1
• explainability 1
• circuits 1
• lab-setup 1
• hands-on 1
• pytorch 1
• interpretability-tools 1
• activation-logging 1
• transformer-lens 1
• tooling 1
• forensics 1
• fingerprinting 1
• clustering 1
• detection 1
• superposition 1
• sparse-autoencoders 1
• features 1
• monosemanticity 1
• visualization 1
• umap 1
• tsne 1
• dimensionality-reduction 1
• concept-maps 1
• causal-tracing 1
• activation-patching 1
• rome 1
• memit 1
• editing 1
• activation-steering 1
• defenses 1
• prompt-injection 1
• interventions 1
• open-source 1
• proposal 1
• architecture 1
• roadmap 1
• series-finale 1

series

Part 11: Steering with Purpose – Predictable Edits, Tested Against Adversaries

14 minute read

This is Part 11 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we captured. In Part 7 we compared. In...

Part 10: Finding the Edit Points – Causal Tracing at the Tensor Level

15 minute read

This is Part 10 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we captured activations. In Part 7 we ...

Part 9: The Concept Cartographer – Mapping Meaning in High Dimensions

14 minute read

This is Part 9 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we captured activations. In Part 7 we f...

Part 8: Untangling Superposition – Reading Features Instead of Neurons

15 minute read

This is Part 8 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we built the activation logger. In Part...

Part 7: The Prompt Fingerprint – Do Similar Prompts Look Similar Inside?

15 minute read

This is Part 7 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we built an activation logger — the tcp...

Part 6: Watching the Brain Fire – Building an Activation Logger for LLMs

14 minute read

This is Part 6 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. We have learned the language of tensors, traced t...

Part 5: Building Your First AI Security Lab – From Theory to Practice

13 minute read

This is Part 5 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. We have spent four articles building the conceptu...

Part 4: Mechanistic Interpretability – Reverse-Engineering the AI Brain

14 minute read

This is Part 4 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. We have covered the math (Part 1), the architectu...

Part 3: The Attack Surface Within – Where Tensors Meet Vulnerabilities

13 minute read

This is Part 3 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 1, we learned what tensors are. In Part 2...

Part 2: How LLMs Actually Think – Embeddings, Attention, and the Math Behind the Magic

13 minute read

This is Part 2 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 1, we built our foundation — understandin...

Part 1: Welcome to the Tensor World – A Security Engineer’s First Steps into Neural Network Mathematics

9 minute read

This is Part 1 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. If you’re a security professional who has ever lo...

Back to Top ↑

llm

Part 2: How LLMs Actually Think – Embeddings, Attention, and the Math Behind the Magic

13 minute read

This is Part 2 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 1, we built our foundation — understandin...

The Ghosts in the Tensors – A New Frontier for InfoSec

2 minute read

In the world of information security, we are comfortable with logic. We understand overflows, injections, and misconfigurations because they follow a traceab...

Back to Top ↑

tensors

Part 1: Welcome to the Tensor World – A Security Engineer’s First Steps into Neural Network Mathematics

9 minute read

This is Part 1 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. If you’re a security professional who has ever lo...

The Ghosts in the Tensors – A New Frontier for InfoSec

2 minute read

In the world of information security, we are comfortable with logic. We understand overflows, injections, and misconfigurations because they follow a traceab...

Back to Top ↑

mechanistic-interpretability

Part 4: Mechanistic Interpretability – Reverse-Engineering the AI Brain

14 minute read

This is Part 4 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. We have covered the math (Part 1), the architectu...

The Ghosts in the Tensors – A New Frontier for InfoSec

2 minute read

In the world of information security, we are comfortable with logic. We understand overflows, injections, and misconfigurations because they follow a traceab...

Back to Top ↑

red-teaming

Part 3: The Attack Surface Within – Where Tensors Meet Vulnerabilities

13 minute read

This is Part 3 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 1, we learned what tensors are. In Part 2...

The Ghosts in the Tensors – A New Frontier for InfoSec

2 minute read

In the world of information security, we are comfortable with logic. We understand overflows, injections, and misconfigurations because they follow a traceab...

Back to Top ↑

embeddings

Part 7: The Prompt Fingerprint – Do Similar Prompts Look Similar Inside?

15 minute read

This is Part 7 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we built an activation logger — the tcp...

Part 2: How LLMs Actually Think – Embeddings, Attention, and the Math Behind the Magic

13 minute read

This is Part 2 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 1, we built our foundation — understandin...

Back to Top ↑

neural-networks

Part 1: Welcome to the Tensor World – A Security Engineer’s First Steps into Neural Network Mathematics

9 minute read

This is Part 1 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. If you’re a security professional who has ever lo...

Back to Top ↑

mathematics

Part 1: Welcome to the Tensor World – A Security Engineer’s First Steps into Neural Network Mathematics

9 minute read

This is Part 1 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. If you’re a security professional who has ever lo...

Back to Top ↑

deep-learning

Part 1: Welcome to the Tensor World – A Security Engineer’s First Steps into Neural Network Mathematics

9 minute read

This is Part 1 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. If you’re a security professional who has ever lo...

Back to Top ↑

attention

Part 2: How LLMs Actually Think – Embeddings, Attention, and the Math Behind the Magic

13 minute read

This is Part 2 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 1, we built our foundation — understandin...

Back to Top ↑

transformers

Part 2: How LLMs Actually Think – Embeddings, Attention, and the Math Behind the Magic

13 minute read

This is Part 2 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 1, we built our foundation — understandin...

Back to Top ↑

adversarial-ml

Part 3: The Attack Surface Within – Where Tensors Meet Vulnerabilities

13 minute read

This is Part 3 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 1, we learned what tensors are. In Part 2...

Back to Top ↑

attack-surface

Part 3: The Attack Surface Within – Where Tensors Meet Vulnerabilities

13 minute read

This is Part 3 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 1, we learned what tensors are. In Part 2...

Back to Top ↑

model-security

Part 3: The Attack Surface Within – Where Tensors Meet Vulnerabilities

13 minute read

This is Part 3 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 1, we learned what tensors are. In Part 2...

Back to Top ↑

reverse-engineering

Part 4: Mechanistic Interpretability – Reverse-Engineering the AI Brain

14 minute read

This is Part 4 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. We have covered the math (Part 1), the architectu...

Back to Top ↑

explainability

Part 4: Mechanistic Interpretability – Reverse-Engineering the AI Brain

14 minute read

This is Part 4 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. We have covered the math (Part 1), the architectu...

Back to Top ↑

circuits

Part 4: Mechanistic Interpretability – Reverse-Engineering the AI Brain

14 minute read

This is Part 4 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. We have covered the math (Part 1), the architectu...

Back to Top ↑

lab-setup

Part 5: Building Your First AI Security Lab – From Theory to Practice

13 minute read

This is Part 5 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. We have spent four articles building the conceptu...

Back to Top ↑

hands-on

Part 5: Building Your First AI Security Lab – From Theory to Practice

13 minute read

This is Part 5 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. We have spent four articles building the conceptu...

Back to Top ↑

pytorch

Part 5: Building Your First AI Security Lab – From Theory to Practice

13 minute read

This is Part 5 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. We have spent four articles building the conceptu...

Back to Top ↑

interpretability-tools

Part 5: Building Your First AI Security Lab – From Theory to Practice

13 minute read

This is Part 5 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. We have spent four articles building the conceptu...

Back to Top ↑

activation-logging

Part 6: Watching the Brain Fire – Building an Activation Logger for LLMs

14 minute read

This is Part 6 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. We have learned the language of tensors, traced t...

Back to Top ↑

transformer-lens

Part 6: Watching the Brain Fire – Building an Activation Logger for LLMs

14 minute read

This is Part 6 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. We have learned the language of tensors, traced t...

Back to Top ↑

tooling

Part 6: Watching the Brain Fire – Building an Activation Logger for LLMs

14 minute read

This is Part 6 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. We have learned the language of tensors, traced t...

Back to Top ↑

forensics

Part 6: Watching the Brain Fire – Building an Activation Logger for LLMs

14 minute read

This is Part 6 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. We have learned the language of tensors, traced t...

Back to Top ↑

fingerprinting

Part 7: The Prompt Fingerprint – Do Similar Prompts Look Similar Inside?

15 minute read

This is Part 7 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we built an activation logger — the tcp...

Back to Top ↑

clustering

Part 7: The Prompt Fingerprint – Do Similar Prompts Look Similar Inside?

15 minute read

This is Part 7 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we built an activation logger — the tcp...

Back to Top ↑

detection

Part 7: The Prompt Fingerprint – Do Similar Prompts Look Similar Inside?

15 minute read

This is Part 7 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we built an activation logger — the tcp...

Back to Top ↑

superposition

Part 8: Untangling Superposition – Reading Features Instead of Neurons

15 minute read

This is Part 8 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we built the activation logger. In Part...

Back to Top ↑

sparse-autoencoders

Part 8: Untangling Superposition – Reading Features Instead of Neurons

15 minute read

This is Part 8 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we built the activation logger. In Part...

Back to Top ↑

features

Part 8: Untangling Superposition – Reading Features Instead of Neurons

15 minute read

This is Part 8 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we built the activation logger. In Part...

Back to Top ↑

monosemanticity

Part 8: Untangling Superposition – Reading Features Instead of Neurons

15 minute read

This is Part 8 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we built the activation logger. In Part...

Back to Top ↑

visualization

Part 9: The Concept Cartographer – Mapping Meaning in High Dimensions

14 minute read

This is Part 9 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we captured activations. In Part 7 we f...

Back to Top ↑

umap

Part 9: The Concept Cartographer – Mapping Meaning in High Dimensions

14 minute read

This is Part 9 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we captured activations. In Part 7 we f...

Back to Top ↑

tsne

Part 9: The Concept Cartographer – Mapping Meaning in High Dimensions

14 minute read

This is Part 9 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we captured activations. In Part 7 we f...

Back to Top ↑

dimensionality-reduction

Part 9: The Concept Cartographer – Mapping Meaning in High Dimensions

14 minute read

This is Part 9 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we captured activations. In Part 7 we f...

Back to Top ↑

concept-maps

Part 9: The Concept Cartographer – Mapping Meaning in High Dimensions

14 minute read

This is Part 9 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we captured activations. In Part 7 we f...

Back to Top ↑

causal-tracing

Part 10: Finding the Edit Points – Causal Tracing at the Tensor Level

15 minute read

This is Part 10 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we captured activations. In Part 7 we ...

Back to Top ↑

activation-patching

Part 10: Finding the Edit Points – Causal Tracing at the Tensor Level

15 minute read

This is Part 10 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we captured activations. In Part 7 we ...

Back to Top ↑

rome

Part 10: Finding the Edit Points – Causal Tracing at the Tensor Level

15 minute read

This is Part 10 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we captured activations. In Part 7 we ...

Back to Top ↑

memit

Part 10: Finding the Edit Points – Causal Tracing at the Tensor Level

15 minute read

This is Part 10 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we captured activations. In Part 7 we ...

Back to Top ↑

editing

Part 10: Finding the Edit Points – Causal Tracing at the Tensor Level

15 minute read

This is Part 10 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we captured activations. In Part 7 we ...

Back to Top ↑

activation-steering

Part 11: Steering with Purpose – Predictable Edits, Tested Against Adversaries

14 minute read

This is Part 11 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we captured. In Part 7 we compared. In...

Back to Top ↑

defenses

Part 11: Steering with Purpose – Predictable Edits, Tested Against Adversaries

14 minute read

This is Part 11 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we captured. In Part 7 we compared. In...

Back to Top ↑

prompt-injection

Part 11: Steering with Purpose – Predictable Edits, Tested Against Adversaries

14 minute read

This is Part 11 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we captured. In Part 7 we compared. In...

Back to Top ↑

interventions

Part 11: Steering with Purpose – Predictable Edits, Tested Against Adversaries

14 minute read

This is Part 11 of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. In Part 6 we captured. In Part 7 we compared. In...

Back to Top ↑

open-source

Part 12: The Bitghost Debugger – A Proposal for Open-Source LLM Instrumentation

14 minute read

This is Part 12 — the final article — of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. Across tensors, transforme...

Back to Top ↑

proposal

Part 12: The Bitghost Debugger – A Proposal for Open-Source LLM Instrumentation

14 minute read

This is Part 12 — the final article — of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. Across tensors, transforme...

Back to Top ↑

architecture

Part 12: The Bitghost Debugger – A Proposal for Open-Source LLM Instrumentation

14 minute read

This is Part 12 — the final article — of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. Across tensors, transforme...

Back to Top ↑

roadmap

Part 12: The Bitghost Debugger – A Proposal for Open-Source LLM Instrumentation

14 minute read

This is Part 12 — the final article — of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. Across tensors, transforme...

Back to Top ↑

series-finale

Part 12: The Bitghost Debugger – A Proposal for Open-Source LLM Instrumentation

14 minute read

This is Part 12 — the final article — of a 12-part series exploring the intersection of artificial intelligence and cybersecurity. Across tensors, transforme...

Back to Top ↑