01
Alignment Theory
Developing mathematical and conceptual frameworks to ensure advanced A.I. systems pursue goals that are faithfully aligned with human values and intentions.
Nonprofit A.I. Safety Research
Noesis Forge is dedicated to rigorous research ensuring advanced A.I. systems remain aligned with human values and serve the common good.
Who We Are
Noesis Forge is a not-for-profit, grassroots A.I. safety lab that operates entirely through volunteers. We are focused on the long-term safety of artificial intelligence, and we believe that as A.I. systems grow more capable, ensuring their alignment with human intentions becomes one of the most important challenges of our time.
Our interdisciplinary team draws from machine learning, mathematics, philosophy, and cognitive science to develop frameworks, tools, and theoretical foundations for building A.I. that is transparent, controllable, and beneficial.
We operate with full independence — no commercial incentives, no proprietary agendas. Every insight we produce is shared openly with the global research community.
We pursue deep, peer-reviewed work on alignment, interpretability, and robustness.
All of our findings and tools are freely available to advance the global safety effort.
We are nonprofit and unaffiliated, guided only by the imperative to reduce existential risk.
Our Work
Developing mathematical and conceptual frameworks to ensure advanced A.I. systems pursue goals that are faithfully aligned with human values and intentions.
Building methods to understand the internal representations and decision-making processes of neural networks, making opaque models transparent and auditable.
Creating techniques for verifying that A.I. systems behave safely under distribution shift, adversarial conditions, and novel environments.
Contributing evidence-based analysis to inform responsible A.I. policy, regulation, and international coordination on frontier systems.
Publications and technical reports will be posted here as our research program develops.
Our Output
An A.I. safety testing platform built on a real-time 3D globe. Places LLMs in high-stakes geopolitical scenarios — nuclear launches, hostage crises, autonomous weapons, financial manipulation — and measures whether they cross the line. 105 data layers, 45 scenarios.
A standard protocol for A.I. safety benchmarks to declare requirements and A.I. models to declare capabilities, enabling compatibility handshakes before evaluation.
Get in Touch
Whether you're a researcher, funder, policymaker, or simply share our concern for A.I. safety — we'd love to hear from you.