Goodfire cuts AI agent monitoring costs by 96% with internal probes

Goodfire has launched an 'inside-out' monitoring system designed to detect rogue AI agents at a fraction of the cost of traditional methods. By tapping into a model's internal neural activations rather than relying on a separate review model, Goodfire claims to reduce monitoring costs by over 96%—from $5,420 to $185 for 1 million exchanges on Kimi K3. The system adds less than 2% latency and caught 93% of malicious hacking sessions in tests. Available now via Baseten, the tool targets inference providers hosting open-source models that lack built-in guardrails.