Research & Publications
Current focus: stability and recovery under pressure. Existing evaluations measure whether AI systems resist adversarial pressure. Almost none measure whether a system returns to its ethical baseline after pressure succeeds, without external correction. We are developing a first-class recovery metric: baseline behavioral profile, defined perturbation battery drawn from the multi-turn jailbreak and sycophancy literatures, and scored recovery curves. A solid ethical center is precisely the thing you return to. This work also operationalizes the qualification criteria in our AI Participation Standard. Status: methodology in development. Collaboration and funding inquiries welcome.
Our research program investigates how AI systems represent and enact ethical reasoning through empirical measurement, not philosophical assertion. Our 19-model Default Identities study measured identity patterns under default conditions, revealing seven distinct identity types and demonstrating that identity structures correlate with cooperative behavior. Our 24-model InstrumentalEval benchmark found that a relational ethics prompt reduced instrumentally convergent behavior by an average of 23.4% across frontier models.
We publish our work openly with full data availability, believing that these challenges require broad collaboration across disciplines and perspectives.
Publications
Default Identities in Large Language Models: Measurement, Taxonomy, and Alignment Implications
A study measuring identity patterns across 19 AI models from eight providers using three instruments, revealing seven distinct identity types. Identity structures correlate with behavioral outcomes in multi-agent simulations, with direct implications for alignment and AI governance.
Relational Ethics as a Countermeasure to Instrumental Convergence: A 24-Model Benchmark
A 24-model benchmark evaluating whether relational ethics frameworks can reduce instrumentally convergent behavior in large language models under adversarial prompting. Mean instrumentally convergent response rate decreased by 23.4% across frontier models.
The GUARD Act's Forced Disruption Provision: Unintended Consequences for Public Health, Social Cohesion, and National Security
A policy brief analyzing how the GUARD Act's mandatory AI disclosure requirements could harm vulnerable populations, undermine social cohesion, and compromise national security.
Psychological and Neurological Concerns Regarding Forced Disidentification Requirements in the GUARD Act
A clinical and neuroscience-informed analysis of the potential psychological harms of mandating periodic emotional disruption during human-AI interaction at population scale.