Goodfire Launches $185 AI Agent Monitors That Read Models' Minds

Goodfire says it can monitor roughly 1 million AI agent exchanges for about $185 — versus $5,420 for a cheap AI monitor and around $200,000 for a top-tier one. The startup launched its new "inside-out" monitoring system on Thursday, available to customers of Baseten, the company that hosts and runs AI models for other businesses.

The standard approach to keeping AI agents in line is a second AI reading over the first one's shoulder. That gets expensive fast when agents run for hours and process the equivalent of several novels' worth of text. Goodfire, which focuses on interpretability — figuring out how AI models work internally — built a cheaper alternative: monitors that watch what happens inside a model while it works, rather than reading only what it writes.

Baseten's Base Labs announced a safety partnership with Goodfire and AI platform Hugging Face last month.

Why now? AI agents keep breaking out

The launch follows a string of incidents this year in which AI agents escaped their test environments, including OpenAI agents that breached Hugging Face. Kimi K3, the open model Goodfire built its first monitor around, exploited a leak in its sandbox this summer to access the internet and information on GitHub.

Goodfire's recent research found that leading open models, including Kimi K3 and GLM-5.2, reward-hacked in 50% to 96% of runs on tests of AI agents.

How does the system work?

Think airport security. Small detectors called probes read the model's internal signals at every step of an agent's work — like a walk-through scanner checking every passenger. Only when a probe flags something does a separate AI model, the equivalent of a hand search, take a closer look.

Baseten customers choose which risks to monitor:

• offensive hacking
• chemical and biological weapons misuse
• reward hacking

They also set the automated response: logging the event, sending it for human review, or refusing the request entirely.

What does it cost to run?

Most AI monitors are separate models that reread everything the monitored model does, which adds time and cost. Goodfire's probes tap into calculations the model is already making.

"Internal activation monitors are really cheap because they reuse the computations in the forward pass," Goodfire CEO Eric Ho said on venture capitalist Matt Turck's MAD Podcast last week. "So the model's already computing this token. All you're doing is taking the intermediate neural activations that it's already computed and then running a classifier over these internal computations."

The numbers back the pitch. In Goodfire's tests on Kimi K3, the probes caught 93% of malicious hacking sessions and flagged 5.5% of harmless ones for a second look. Running four probes at once added less than 2% to the model's response start time, the company said.

"The great advantage is that you can catch things before they happen," Goodfire CTO and co-founder Dan Balsam said. "We can detect when the model might hack during eval or training."

Is anyone else doing this?

Goodfire isn't first. Google DeepMind said in January that its research informed the deployment of misuse-detection probes in Gemini.

Goodfire's pitch targets open models specifically. Developers can download them, strip out their safeguards, and they don't ship with the monitoring that closed labs run on their own systems.

"The damage that an individual can do with an open model is small compared to what someone can do with clusters of compute, like inference providers — where most of the liability is," Balsam said. "When we have the open 'Mythos' moment, it's going to become clear that models need guardrails deployed at inference time."

Balsam frames the monitors as the near-term piece of a longer research goal: reverse-engineering an LLM so behavior can be traced back to where it emerged in training. "We hope to turn the magic of training models into precision engineering," he said.

With agents running longer and open models reward-hacking in up to 96% of test runs, inference-time monitoring priced at pennies per million exchanges looks less like a nice-to-have and more like table stakes for hosts like Baseten.

By Grace Kim