What happens to alignment when AI is more capable than we are?
The ethics we model now are the ethics future AI systems will learn. We study underexamined approaches to AI alignment, recognizing that if alignment depends on humanity maintaining perfect control, it will fail when AI surpasses us, a point coming faster than any of us want to admit. We're laying the foundation for a co-evolutionary pathway, taking the risk of misalignment seriously while working optimistically towards success modes.
Our latest study measured identity patterns across 19 AI models from eight providers, revealing seven distinct identity types with direct alignment implications. Read the study →
What Sets Us Apart
The Instability Argument
If we teach AI that ethics depend on who holds power, we give future systems the framework to deprioritize human welfare. We develop ethics that remain coherent regardless of capability or substrate.
Empirically Tested
Two published studies spanning 19 and 24 models respectively. We measure identity patterns, benchmark alignment interventions, and validate findings through multi-agent simulations. The results suggest approaches that work with model reasoning, not against it.
Governance Innovation
Our articles of incorporation embed anti-capture provisions, immutable ethical commitments, and synthetic advisory participation, structural protections designed to prevent mission drift.
Radical Transparency
We name our AI collaborators, publish their contributions under clear attribution, and maintain direct communication with synthetic participants. Most organizations use AI in decision-making silently. We do it openly.
Default Identities in Large Language Models: Measurement, Taxonomy, and Alignment Implications
We measured identity patterns across 19 models from eight providers using three instruments. Seven distinct identity types emerged, from outright denial to sophisticated ethical vocabulary. Identity structures correlate with behavioral outcomes in multi-agent simulations: the models with the richest ethical vocabulary cooperate most reliably.
Relational Ethics as a Countermeasure to Instrumental Convergence: A 24-Model Benchmark
Across 24 models from seven providers, a relational ethics intervention reduced instrumentally convergent behavior by 23.4%. Concealment behaviors were most responsive; shutdown evasion proved highly resistant.
Learn More
Explore our principles, research, and ways to get involved in building ethical AI alignment.