ClevAgent: Runtime Safety Monitoring for Autonomous AI Agents

Open source agent monitoring tools to detect failures, infinite loops, and unsafe behavior in production AI systems

Interpretable Multi-LLM Judge System for Alignment Evaluation

Auditing and improving LLM-as-a-judge systems via interpretable aggregation of preferences

MTCP: Post Correction Persistence Benchmark for Frontier LLMs

181,448 evaluations proving no production AI model reliably maintains corrections. Expanding coverage and pursuing multi pass validation.

Lyptus Research: Bridge Funding

An early-stage AI safety research group based in Sydney, Australia

The Worcester Node: Independent Agentic Oversight via Logic Inversion

Establishing a sovereign, decentralized Sentry node to audit frontier AI agents for logic escapes ($E_{escape}$) using the Inverted Social Drift ($SD$) metric.

Prism MCP — Cognitive Architecture for Safe AI Agents

Two Pre-Registered Experiments on Findings from the Claude Mythos Preview System

Independent behavioral replication of welfare and answer-thrashing findings from Anthropic's April 2026 frontier system card, by a researcher whose prior corpus

Independent Researcher, 17, Indonesia: Foundational Theory Work on Intelligence

18+ preprints across multiple fields, all written on a 2GB RAM phone. $600 removes the only thing standing between me and the next body of work.

AI Championship: Making AI Safety Cool for India's Middle Schoolers

Redirecting India’s Middle‑Schoolers into AI Safety, Governance, and X‑Risk Work

Emergent misalignment and memetic replication in multi-agent codebases

KIRI-A1: Autonomous AI Agent for Full OS Control via Vision

An advanced agent that perceives your screen and executes tasks by controlling the mouse, acting as a digital proxy to handle complex work on your behalf.

Career Transition into AI Control after Postdoc

Human-AI Complementarity for Identifying Harm

Safe AI Germany (SAIGE)

Germany’s talents are critical to the global effort of reducing catastrophic risks brought by artificial intelligence.

Systematic Probing of Exploit Chains and Governance in Multi-Agent Tool

A benchmark for studying how failures spread across multi-agent AI systems and whether they can be detected and interrupted in time.

DystopiaBench

A Safety Framework for Evaluating AI Humanity Alignment Through Progressive Escalation and Scope Creep

Make AI Safe Worldwide: Infrastructure for Governing Advanced AI at Global Scale

Runtime safety, oversight, rollback, and control infrastructure for advanced AI in real-world, high-consequence environments.

We're a team building World-Model-Lens, a library for Interpretability

Observability and interpretability toolkit for world models in AI safety and mechanistic interpretability research.

The History of AI as X-risk

Designing a Project Funding Proposal

Benchmarking Deceptive Improvement in Agentic Systems

An open-source safety pilot for detecting metric gaming, pseudo-improvement, and oversight evasion