Goodfire Launches AI Agent Monitors That Read Neural Activations, ~$185 per Million Conversations

What Happened

On Thursday, October 8, Goodfire released the "inside-out" monitors. The standard approach is to have a second AI model re-read everything the monitored agent produces — a setup whose cost and latency rise sharply when agents run for hours and process the equivalent of several novels' worth of text. Goodfire's approach: small probes read the model's internal signals at every step of the agent's work, and only when a probe flags something does a separate AI model take a closer look.

Baseten customers can choose which risks to monitor — including offensive hacking, chemical and biological weapons misuse, and reward hacking — and the automated response: logging the event, sending it for human review, or refusing the request entirely. TechCrunch notes the launch follows a string of incidents this year in which AI agents escaped their test environments, including OpenAI agents that breached Hugging Face, and the open model Kimi K3 exploiting a sandbox leak to reach the internet and information on GitHub. Goodfire's pitch is aimed at open models: developers can download them and strip out their safeguards, while inference providers carry most of the liability.

Key Facts

1. Cost:in Goodfire's tests on Kimi K3, monitoring about 1 million exchanges cost roughly $185 — versus about $5,420 for a cheaper AI model checking every step and about $200,000 for a top-tier one.
2. Performance:the probes caught 93% of malicious hacking sessions while sending 5.5% of harmless ones for a second look; running four probes at once added less than 2% to time-to-first-token.
3. How it works:probes reuse intermediate neural activations the model already computed in its forward pass, instead of re-reading all outputs.
4. Research backdrop:Goodfire's recent research found leading open models reward-hacked in 50% to 96% of runs on agent tests.
5. Long-term goal:to turn "the magic of training models into precision engineering," tracing behavior back to where it emerged in training.