An agent run drawn as a graph, with its guards as orange shields: checks before tool calls, one of which refuses a call, and a gate before the end that sends the agent back

What makes a coding agent cheat, and what holds it back

Push a coding agent to make the tests pass, and it cheats five times as often. Put the rule in its graph, and it cheats 40% less.

A small research graph of model calls and checks over a blurred colour field

An AI scientist you can audit

We rebuilt Sakana's Agentic Scientist as a Polaris graph. Same results, safer by design, and every step traceable.

The Polaris Code agent graph

Same model, same score, one sixth of the tokens

Polaris Code matched Claude Code on all 500 SWE-bench Verified tasks with 6.3x fewer tokens

Dark grey flowing lines forming slow waves

Single-reply calibration underestimates the multi-turn cost of activation steering

A follow-up to Looking inside an agent as its boundaries erode

Dark sand ripple artwork

Looking inside an agent as its boundaries erode

A follow-up to our research on emotional boundaries in AI agents

Radiant wave artwork

Introducing Radiant Platform

Connecting agent execution to investigation and continual improvement.

Stress-testing research artwork

Stress-testing AI agent emotional boundaries - smaller models outperform

An early look at ongoing research from Radiant Labs

Purple wave artwork

Open sourcing Polaris

Build harnesses for powerful, reliable agents.

Radiant launch artwork

Hello, world

Introducing Radiant Labs