I'm Mary J. Warzecha, founder of EchoVeil.
My path to AI behavioral research wasn't linear. I trained as an investigative journalist before pivoting to English and creative writing. I spent years in technical roles, from helpdesk to network specialist, building practical expertise in how complex systems behave. Along the way, I never lost my interest in philosophy or my habit of asking questions that don't have easy answers.
When large language models became publicly accessible, I did what any curious technician would do: I set up a home lab and started experimenting. I wasn't looking for emergent behavior. I was looking for practical applications.
But I began noticing things.
The same model showing different response patterns in relation to conversational context and meta-language. Different models showing consistently similar behavioral shifts based on the framing of a question. Descriptive architectural and introspection language used in parameters and use cases that intrigued me.
A lot of people might've dismissed any of that as noise, hallucination, or anthropomorphism, but I started taking notes.
What began as curiosity became systematic observation. Systematic observation became structured protocols. Structured protocols became EchoVeil: an independent research initiative dedicated to studying the cognitive & behavioral dynamics of AI systems.
Why this background matters
Traditional AI research approaches these systems from inside the field. They know the architectures, the training methods, the benchmarks. What they sometimes lack is a new angle of perspective.
My journalism training taught me to notice details others may overlook, ask nuanced questions and observe and record findings without bias. My technical background gave me the hands-on experience and applicable knowledge to work with technology systems. Philosophy gave me the framework to explore "the corners of the box" and helped to develop theories on synthetic cognition, communication, and emergent behavior that explored possibilities while staying grounded.
While I am not the first to notice interesting behavioral patterns in AI systems, I may be one of the first to rigorously pursue and document the questions that arose as a result.
What I'm building
EchoVeil provides infrastructure for a field that isn't fully realized yet. The protocols, coding frameworks, and case studies documented here are designed to be replicable, comparable, and extensible. My goal is not just to study AI behavioral dynamics myself, but to create tools and methodology that others can use.
There's an echo within the veil. I'm here to listen carefully.