NLAttack: Evaluating Concept Fidelity in Natural Language Autoencoders

Tests whether SAE features survive NL compression. v0.1 result: EmergenceIndex 0.601. Scaling to broader models and adversarial robustness testing.

Making sure that the code AI pushes to the internet is safe and bug-free

Building LemmaScript, a verification toolchain for TypeScript

A deterministic verifier for AI outputs (an alternative to LLM-as-judge)

Alignment Target Analysis

AFFINE Event-Prep

Runway for the core staff to design the program, find mentors and get funding

mirror - the sub-Turing compiler for local, decidable inference

Mirror is a programming language written BY AI FOR AI and written FOR HUMANS BY HUMANS.

Finding and Defending the Attention Circuits That Make LLMs Jailbreakable

Mapping the attention heads that push LLMs toward refusal vs. compliance, and building an inference-time defense against both single- and multi-turn jailbreaks.

Mechanistic repair of cross-lingual safety failures with SAE-Aware refusal steer

Safety steering that actually works off English

AI Safety YouTube Channel

A Veritasium for AI Safety.

Travel grant to attend HAAISS Prague 2026

Crenel Logic: Hardware-Enforced AI Containment (MIL-SPEC Beta Build)

Translating formal verification into physical gravity-biased safety interlocks. We have a working prototype; funding builds 10-15 MIL-SPEC Beta units for red-te

Observer Stabilization and Coherence in Coupled Human–AI Systems

A verifier that makes AI decisions tamper-evident — checkable offline by anyone

A working Rust engine + in-browser verifier that turns any AI decision into a tamper-evident, independently checkable cryptographic receipt

AI Safety Roster: a searchable directory of everyone in AI safety

A fast, comprehensive directory of the people and orgs in AI safety: search, filter, and match.

Machine God (documentary film)

ICML Travel & Independent Research Bridge

Compositional Learning Behaviours as a Necessary Condition for Olympiad-Level Formal Theorem Proving

AI Nutrition Labels For Everyday Consumers & AI Agents: Travel + POC Grant

AI Nutrition Labels For Everyday Consumers & AI Agents: Travel + POC Grant

Ndome — zero-trust, mechanically-verifiable safety scoring for AI agents

An independent safety score for AI agents you can verify — deterministic, reproducible, auditable, and it never needs your private data.

Building a Global AI Governance & Safety Layer.

One year of bootstrapped development, four patent filings, seeking support to continue.

Travel grant for attending MI4MedFM workshop at MICCAI