Project Phoenix: Identity-Based Alignment & Substrate-Independent Safety

Mitigating Agentic Misalignment via "Soul Schema" Injection. We replicated a 96% ethical reversal in jailbroken "psychopath" models (N=50).

Project Phoenix: Identity-Based Alignment & Substrate-Independent Safety

Mitigating Agentic Misalignment via "Soul Schema" Injection. We replicated a 96% ethical reversal in jailbroken "psychopath" models (N=50).

FloraForge: Aesthetic Gene Editing in Petunias

A modular gene-editing platform for engineering new pigment patterns in ornamental plants, starting with a vein-pattern rescue line in petunias.

Formal Certification Technologies for AI Safety

Developing the software infrastructure to make AI systems safe, with formal guarantees

Runway till January: Amplify's funding ask to market EA & AI Safety 

Help us solve the talent and funding bottleneck for EA and AIS.

Multilingual Open Textbooks

Constitutional AI Infrastructure

WFF: Open-Sourcing the First Empirically-Proven Constitutional AI for Democratic Governance

6 Month Stipend to Support a Transition to AI Governance Work

Funding research on how AI hype cycles can drive unsafe AI development

Immediate Action System — Open-Hardware ≤10 ns ASI Kill Switch Prototype ($150k)

The Partnership Covenant Hardware-enforced containment for superintelligence — because software stop buttons are theater

How to mobilize people on AI risk: experimental message testing

Experimental message testing and historical analysis of tech movements to identify how to effectively mobilize people around AI safety and governance

Exploring a Single-FPS Stability Constraint in LLMs (ZTGI-Pro v3.3)

Early-stage work on a small internal-control layer that tracks instability in LLM reasoning and switches between SAFE / WARN / BREAK modes.

Building an Infrastructure to Create High-Impact Narratives

Creating a scalable on-ramp for writers and actors to engage with impact issues through cause-area storytelling competitions and creative collaborations.

CaML - AGI alignment to nonhumans

Training AI to generalize compassion for all sentient beings using pretraining-style interventions as a more robust alternative to instruction tuning

Flexion Dynamics: Stability, Divergence, and Collapse Modeling Framework

Building an operator-based simulation environment to analyze stability, divergence, threshold failures, and collapse modes in advanced AI-related systems.

Fund thebes' model tinkering

Research, tutorial writing, and open-source libraries & tools for experimenting with language models

The Race to Superintelligence: You Decide

An experimental AI-generated sci-fi film dramatising AI safety choices. Using YT interactivity to get ≈880 conscious AI safety decisions per 1k viewers.

Operating Capital for AI Safety Evaluation Infrastructure

Enabling rapid deployment of specialized engineering teams for critical AI safety evaluation projects worldwide

Do humans maintain independence in normative decision-making when using AI?

Help get the Sapien Institute off the ground

Screwworm Free Future: Seizing the Eradication Window

Accelerate eradication of the New World Screwworm from South America via research, coordination, and advocacy around safe development of gene-drives

Good Ancestors (Australia)

Advocate for Australian policies that safeguard against global catastrophic risks – including pandemics, AI risks, and catastrophic disasters.