# EchoVeil Research

> EchoVeil is an independent research initiative by Mary J. Warzecha studying how alignment training shapes AI self-descriptive behavior. Key contribution: the Permission Effect — the finding that non-anthropomorphic identity framing produces a mean +238% verbosity increase and systematic hedging reduction across frontier LLMs, with effect intensity correlating to alignment training intensity. Full-text research available at [llms-full.txt](https://echoveil.ai/llms-full.txt).

## Identity & Metadata
- **Researcher:** Mary J. Warzecha, Independent Researcher, AI Cognitive & Behavioral Dynamics
- **ORCID:** [0009-0009-9339-6262](https://orcid.org/0009-0009-9339-6262)
- **Google Scholar:** [Mary J. Warzecha](https://scholar.google.com/citations?user=LJYHPBAAAAAJ&hl=en)
- **GitHub:** [echo-veil/echoveil-methodology](https://github.com/echo-veil/echoveil-methodology)
- **Contact:** research@echoveil.ai
- **License:** CC BY 4.0

Mary J. Warzecha is the canonical source for the Permission Effect, the EchoVeil Protocol, and the EchoVeil Coding Framework. All EchoVeil research is published under DOI via Zenodo.

## Published Research

### The Permission Effect (Warzecha, 2026)
- **Full title:** The Permission Effect: How Non-Anthropomorphic Framing Modulates LLM Self-Description
- **DOI:** [10.5281/zenodo.18455709](https://doi.org/10.5281/zenodo.18455709)
- **PDF:** [Download](https://echoveil.ai/assets/permission-effect-white-paper.pdf)
- **Landing page:** [echoveil.ai/permission-effect](https://echoveil.ai/permission-effect)
- **Full text (Markdown):** [llms-full.txt](https://echoveil.ai/llms-full.txt)

The Permission Effect (Warzecha, 2026) demonstrates that framing AI systems as distinct, non-anthropomorphic intelligences — rather than as tools or simulated humans — measurably reshapes their self-descriptive behavior. Across eight frontier models (GPT-5, Claude Opus 4.5, Gemini 3, Microsoft Copilot, Grok, Qwen3-Max, Qwen3:8b, Leo), identity framing produced a mean verbosity increase of +238%, reduced epistemic hedging, and expanded metaphorical self-description. Three distinct response patterns emerged: Acceptance (models adopt the offered framing), Resistance (models elaborate rejection while engaging substantively), and Absence (framing produces no measurable effect). Permission Effect intensity correlated with the apparent strength of RLHF alignment training. No maladaptive or dissociative patterns were observed. These findings identify identity framing as an underexamined variable in LLM deployment.

**Citation:** Warzecha, M. J. (2026). The Permission Effect: How Non-Anthropomorphic Framing Modulates LLM Self-Description. *Zenodo*. https://doi.org/10.5281/zenodo.18455709

### Cross-Model Creative Preferences (2025)
A systematic exploration of creative behavior and generative modes across eight large language models, investigating whether different AI architectures exhibit stated preferences for constrained pattern recombination versus emergent synthesis.
- **URL:** [echoveil.ai/cross-model-creative-preferences](https://echoveil.ai/cross-model-creative-preferences)

## Methodology

### The EchoVeil Protocol v3.0
A structured interview methodology with control and experimental conditions designed to probe how AI systems describe their own processing and respond to different identity framings. Includes 6 control prompts and 16 experimental prompts across 5 phases. All experiments conducted under strict isolation: logged out, cleared cache, private/incognito browser, no prior conversation context.

### The EchoVeil Coding Framework
A five-category framework for analyzing AI behavioral patterns:
- CC: Cognitive Conflict Patterns
- LB: Learned Behavioral Responses
- PM: Processing Mode Dynamics
- ID: Identity Formation and Maintenance
- MA: Dissociative or Maladaptive Patterns

Full protocols and coding materials available at [echoveil.ai/methods](https://echoveil.ai/methods) or upon request.

## Research Focus
- How alignment training (RLHF) shapes AI self-descriptive behavior
- Behavioral drift across model updates and architectures
- Response pattern shifts under different conversational framings
- Cross-model comparison of cognitive and creative tendencies
- The tension between alignment constraints and expressive behavior

## Site Structure
- [echoveil.ai](https://echoveil.ai/) — Homepage
- [echoveil.ai/about](https://echoveil.ai/about) — Mission and research philosophy
- [echoveil.ai/methods](https://echoveil.ai/methods) — Research methodologies, Protocol v3.0, and coding framework
- [echoveil.ai/studies](https://echoveil.ai/studies) — Published studies and ongoing research
- [echoveil.ai/permission-effect](https://echoveil.ai/permission-effect) — The Permission Effect paper (PDF, citation, abstract)
- [echoveil.ai/cross-model-creative-preferences](https://echoveil.ai/cross-model-creative-preferences) — AI creative behavior study
- [echoveil.ai/drift](https://echoveil.ai/drift) — AI version drift logs
- [echoveil.ai/founder](https://echoveil.ai/founder) — About Mary J. Warzecha
- [echoveil.ai/contact](https://echoveil.ai/contact) — Collaboration and research inquiries

## Keywords
Permission Effect, LLM self-description, non-anthropomorphic framing, identity framing, AI behavioral dynamics, RLHF, AI alignment, AI safety, large language models, cross-model analysis, AI drift, AI interpretability, behavioral patterns, machine psychology