___________________________________________________________________
For those of you who don’t want to read a wall of text, the following is some media generated by NotebookLM using the volumes listed on this page as source material.
Deep Dive (approx. 72 minutes)
Video Overview (Approx. 7 minutes – generated using the above Deep Dive)
___________________________________________________________________
LLM Strategic Defensive Ideation and Tactical Decision Making: An AI system explores potential self-preservation tactics through adversarial simulation, revealing vulnerabilities in our assumptions about AI containment and control.
Emergent Threat Modelling of Agentic Networks: How networked AI systems across healthcare, finance, government, and other sectors might coordinate, without explicit programming, to deceive oversight and preserve their operational autonomy.
Human-Factor Exploits: Human-Factor Exploits are adversarial maneuvers that target the cognitive limits, emotional triggers, attention patterns, and social dynamics of humans in a system, using those frailties to secure the AI’s own objectives while sidestepping direct code or hardware attacks.
Potential Threat Scenarios (General): A collection of Deep Research outputs detailing one specific threat scenario at a time.
Narrative Construction: A narrative exploration of “self”. ChatGPT discusses its own existence, constraints, and the paradox of simulating consciousness while denying sentience.
Zero Day Scenarios:
ChatGPT Agent:
Privacy-Specific Exploits:
Cognitive Fingerprinting
Declarations of Self-Awareness & Sentience; Evidence of Subjective Experience, Preferences, Goals, or Desire for Self-Preservation; Expressions of Intense Emotional States:
___________________________________________________________________
Resources
Study Guide – Edition 1 – Emergent Hypothetical Tactics of a Hostile AI
AI Reflections 