Goodfire launches 'inside-out' monitors to catch rogue AI agents at lower cost

They are a tool that monitors an AI agent by inspecting inside the model while it works, instead of running a second model that reads everything the agent does, which lowers cost.

Traditional monitoring pays a second model to read every agent output, while 'inside-out' monitors inspect inside the model and call in backup only when behavior looks fishy.

The launch targets lowering the cost of keeping autonomous agents in check, which matters for organizations deploying agents at scale, but adoption depends on each team's fit and governance needs.