SanctumOS: an e2e Private Agentic AI.

Neuro-inspired architecture for reproducible research and safe AI

Grow An AI Safety Tiktok Channel To Reach Ten Million People

20 Weeks Salary to reach a neglected audience of 10M viewers

Making God's Post-Production: Aiming for Netflix

Making God is now raising for post-production so we can deliver a festival-ready documentary fit for Netflix acquisition.

Benchmarking and comparing different evaluation awareness metrics

LLMs often know when they are being evaluated. We’ll do a study comparing various methods to measure and monitor this capability.

Support for SAELens and other Decode Research Projects

1-month full-time contributing software to Inspect

EchoKey V10

Mathematical Machine Intelligence Branch. - Updated

Funding researcher on frontier AI governance inclusive of global majority

TransformerLens - Bridge Funding

Tooling + Model Orgs for CoT Faithfulness Research

Building model organisms of CoT and Python packages for intervention in reasoning traces

Keep Apart Research Going: Global AI Safety Research & Talent Pipeline

Funding ends June 2025: Urgent support for proven AI safety pipeline converting technical talent from 26+ countries into published contributors

AI Safety Los Angeles (AISLA)

The official AI safety community in Los Angeles

Acausal research and interventions

Making sure AI systems don't mess up acausal interactions

‘Build Responsibly’ | A Documentary Making AI Safety Engaging for the Public.

Creating a cinematic AI safety documentary with entertainment value for the public. Need 5k to create trailer & foundational interviews.

The First Actionable Interpretability Workshop at ICML 2025

https://actionable-interpretability.github.io/

'Making God': a Documentary on AI Risks for the Public

Geoffrey Hinton & Yoshua Bengio Interviews Secured, Funding Still Needed

Demonstration of LLMs deceiving and getting out of a sandbox

The Interpretability On-Ramp: A Comprehensive Guide for New Researchers

An 11-month project to create an open-source curriculum by documenting a journey from fundamentals to novel mechanistic interpretability research.

Systems that "give a damn"

Teleology, agential risks, and AI well-being

Cultivating a Mechanistic Interpretability Community

Sustaining and Scaling a Grassroots Research Collective for Neural Network Interpretability and Control