Scenario design
Scenarios that probe dispositions such as cooperation, deception, and resource acquisition, varied across agent count, model, and autonomy.
AI safety research
Studying AI agents, together.
Akxar Labs is an early-stage research lab building open environments and oversight tools to study how AI agents behave, individually and in groups.
Scenario designEnvironments that probe specific behavioral dispositions. Multi-agent oversightMonitoring and control for agent systems at scale. Open infrastructureShared environments researchers can build on.
Featured research
Our first releases.
SwarmFrame: tools for overseeing large agent swarms
A customizable dashboard and sandbox for building and testing oversight methods for systems with thousands of agents.
Newts' Lab: a testbed for overseeing automated research
A customizable lab where an AI agent runs research from idea to paper under human gates, built for studying how such systems should be overseen.
Research directions
Evaluating agents in open-ended settings.
Open-ended multi-agent environments surface behavior that short, single-agent evaluations miss: long-horizon coordination, mutual influence, and unexpected strategies. We are building the tools to study it.
Oversight and monitoring
Interfaces for observing and steering multi-agent simulations, and a testbed for comparing oversight and control mechanisms.
Shared infrastructure
Common environments and monitoring tools that other researchers can use with their own questions and agents.