Goodfire Launches Inside-Out Monitors for Lower-Cost AI Oversight

What changed

Goodfire launched an “inside-out” monitoring system for AI agents. It watches what is happening inside the model as the agent works, then asks a second AI to investigate only when behavior looks suspicious, instead of having that second system review everything.

Why it matters

The design targets one of the awkward costs of agent oversight: constant supervision can require running another AI alongside the first one. Selective checks could make monitoring cheaper while keeping a backup reviewer available for unusual behavior.

That changes the practical question for AI-agent operators. They may not need to choose between expensive continuous review and little visibility at all. Goodfire says its approach costs less, but the reduction is not quantified, and the reporting so far does not establish how accurately or quickly the monitors detect problems, which models they support, or whether customers are using them.