Database

I self-funded research into a new threat model. It is demonstrating impact (accepted at multiple venues, added to BlueDot's curriculum).

Compute Funding

Fund a new research agenda, based on computational mechanics, bridging mechanism and behavior to develop a rigorous science of AI systems and capabilities.

We are fostering the next generation of AI Policy professionals through the Talos Fellowship. Your help will directly increase the number of places we can offer

An association for interdisciplinary interest in AI 

Empowering everyone to detect and combat AI-generated content threats with advanced multi-modal verification tool

An audit-grade evaluation of persistent influence, reset failure, and isolation assumptions in long-context AI systems

A trusted profession that has advocated against existential risks like nuclear war can do so again for AI — but clinicians must first be made aware of the risks

Automated creation of defensive tools like AI control protocols and defensive cybersecurity agents

The Official AI Safety Community in Los Angeles

Extending an AI control evaluation to include vulnerability discovery, weaponization, and payload creation

Funding top-up for an early-career reseacher to attend Global Challenges Project (GCP) Workshop for career exploration in mitigating GCRs

344 MIT rules merged into Microsoft Agent Governance Toolkit, Cisco AI Defense, MISP, OWASP. Microsoft Copilot SWE Agent uses ATR for CVE triage.

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

No summary available yet.

6 months of work: Evaluating a variant of GPT2-XL that can simulate a shutdown activation, aiming to improve alignment theory & develop interpretability tools.

A Research Agenda for Sovereign Capability

A flexible simulation environment for assessing strategic and persuasive capabilities, benchmarking, and agent development, inspired by reality TV competitions.

No summary available yet.

Proves observed alignment under monitoring ≠ intrinsic policy. Full simulator, 1,000-scenario audit, and general theory of entity freedom (ϕ_x).

Addressing Immediate AI Safety Concerns through DevInterp

Seeding a business which finds grants and High Net Worth Individuals beyond EA

No summary available yet.

by buying gift cards for the game and handing them out at the OpenAI offices

No summary available yet.

Benchmark for agent safety when spending users money. How often do they violate user intent and rules?

Surveying neuroscience for tools to analyze and understand neural networks and building a natural science of deep learning

No summary available yet.

The first AI safety evaluation benchmark for Nigerian indigenous livestock systems testing whether frontier models are safe to deploy in African food systems.

This round of funding will be used primarily for prototype hardening, artifact packaging, runtime evaluation, and preparation for external review.

Advocating for U.S. federal AI safety legislation to reduce catastrophic AI risk.

I've self funded my ramp up for six months and interview/grant processes are taking longer than expected.

Promoting better management of Global Catastrophic Risks in Spanish-Speaking countries.

No summary available yet.

No summary available yet.

Organizing global AI ethics think tank for dynamic AI research updates and framework for AI safety policies implementation and humanity income support

Running the initial online version of a 4-week biosecurity course for 20-50 participants

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

Social media content across YouTube, Instagram, and TikTok to grow AI x-risk awareness and build political momentum for a global pause.

Inspiring India’s Middle‑Schoolers to pursue AI Safety, Governance, and X‑Risk Work

No summary available yet.

Developing an innovative wisdom layer for AI that enhances its capabilities for deep analysis, safe AI, and creative solutions to complex systemic problems.

De-risking AI Catastrophe: A cyber-physical protocol using ZKPs and NIR Spectroscopy to resolve the governance deadlock in critical global infrastructure.

Developing correct-by-construction world models for verification of frontier AI

No summary available yet.

We build a scalable "Automated Circuit Discovery" method and investigate "Cleanup Behavior" to advance the interpretability of transformer models.

Measured post-embodied sensation integration. Solo daily-pace brain-function development in Osaka. Phase 1 funds higher cognitive integration program.

Triadic geometric training data and architecture replaces RLHF

Six-month support for a Program Manager to organize and execute international AI safety hackathons with Apart Research