Rewind: Atomic Rollback Infrastructure for AI Agents

Ctrl+Z for AI agents. Rewind automatically snapshots the filesystem and the conversation together, so when your agent breaks something, one rollback fixes both.

ValiChord: verification for AI evals and science that can't be gamed

Independent validators test a claim blind, seal their verdicts, and reveal together — a permanent, tamper-proof record of AI evals and scientific results.

Postdoc on verification protocols for frontier AI training

Low-overhead zero-knowledge proofs of properties of training

A neutral and unbiased AI governance and safety system for the world

AI safety which operates on both model/agent and global level - a unique solution as far as we know.

An Open Authority Standard for Embodied AI Agents That Act on the Physical World

An open schema, consent framework, and harm taxonomy bounding what embodied AI agents may do in the home, validated on a shipping humanoid robot.

What breaks when a robot never stops learning? Continual RL on real hardware

A Dreamer 4-style multimodal world model with a grounded, self-generated reward, learning 24/7 on a $500 open arm — measuring wireheading, forgetting, drift

ChronoMirror

Deployment-Time Interception of RLHF Alignment Failures

Theoretical research of inductive reasoning building on LTA

COMPASS Fellowship

Capacity-Oriented Mentorship for Public Administrators on AI Safety & Strategy

Testing failure modes of debate-style AI control schemes

Concept control in transformers with Sparse Concept Anchoring

Develop a training-time method for transformers that puts concepts where you can find them, so removal has predictable efficacy and bounded side-effects.

Grant to establish an AI safety lab and fellowship from India

Focusing more on the intersection of red teaming and interp, while creating a strong rooted community.

Project CHRONOS: Ephemeral AI with verifiable cryptographic self-destruct

6-month stipend and compute to harden an open-source FHE and VDF architecture for secure, time-bound autonomous agents

ClauseHound

Indus AI Node

Mechanistic estimation of highly out-of-distribution behaviors

AI & The City: Tiktok Street Interviews

An AI social media project from a diverse content creator who has accumulated over 100k followers on 2 separate accounts

Emotion as Attack Surface and Monitoring Signal in Thinking Models

Debate training on LLMs as a reward-hacking mitigation

Diamond Engine: Contextual Attention Perturbation for Intentional Hallucination

"Inducing context-driven chaos during inference to unlock non-linear token pathways, resolved through autonomous semantic homeostasis."

Discrete Language Modeling