Goodfire Launches Model Internal Activation Monitor for Rogue Agents
Understand the story
Interpretability startup Goodfire has launched a monitor for AI agents. Unlike external AI systems that read model outputs, the monitor reads signals directly inside the model, and is now available to Baseten customers. In tests against the Kimi K3 model, monitoring about 1,500 sessions cost roughly $51—significantly less than the $233 cost of step-by-step checks with a cheap AI model, and far below the roughly $10,000 cost of using a top model. On performance, the monitoring probes captured 94% of malicious intrusion sessions with a false positive rate of 8.7% on harmless sessions, and running four probes in parallel added less than 2% response latency.
AI overview of reporting · Updated 10 days ago
Reporting timeline
Explore different aspects of the story through its reporting.
- TechCrunch · AIGoodfire says its new ‘inside-out’ monitors catch rogue AI agents at a fraction of the cost
Interpretability startup Goodfire has launched a monitor that reads signals inside the model rather than having an external AI re-read outputs, now available to Baseten customers. In tests on Kimi K3, monitoring about 1,500 sessions cost roughly $51, far below $233 for step-by-step checks by cheap AI models and about $10,000 for top models; the probes caught 94% of malicious intrusion sessions with an 8.7% false-positive rate on harmless sessions, and running four probes in parallel added less than 2% response latency.