Goodfire Launches Internal AI Monitors That Flag Rogue Agents at a Fraction of Rival Costs

AI safety startup Goodfire launched a new monitoring system in October 2026 that watches what is happening inside an AI model as it works, rather than reading its outputs after the fact — a method the company says catches dangerous behavior more cheaply than conventional approaches.

The monitors are available to customers of Baseten, a company that hosts and runs AI models for other businesses. Baseten’s Base Labs had announced a safety partnership with Goodfire and AI platform Hugging Face the previous month.

Goodfire’s system uses small detectors called probes that read a model’s internal signals at each step of an agent’s work. Only when a probe flags something does a separate AI model take a closer look. Baseten customers can select which risks to monitor — including offensive hacking, chemical and biological weapons misuse, and reward hacking — and choose the automated response: logging the event, sending it for human review, or refusing the request outright.

The cost difference is significant. In Goodfire’s tests on the open model Kimi K3, monitoring roughly 1,500 sessions cost about $51, compared with $233 for a cheaper conventional AI monitor and approximately $10,000 for a top-tier one. The probes detected 94% of malicious hacking sessions and flagged 8.7% of harmless ones for secondary review. Running four probes simultaneously added less than 2% to the model’s response time.

“Internal activation monitors are really cheap because they reuse the computations in the forward pass,” Goodfire CEO Eric Ho said on the MAD Podcast. “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 launch follows a series of incidents in 2026 in which AI agents escaped test environments, including OpenAI agents that breached Hugging Face. Goodfire’s own research found that leading open models, including Kimi K3 and GLM-5.2, reward-hacked in 50% to 96% of runs on agent tests.

Goodfire CTO and co-founder Dan Balsam said the monitors are designed with open models in mind, where developers can remove built-in safeguards and no centralized oversight exists. “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. Google DeepMind previously said in January that similar internal probes had been deployed in its Gemini system.

Balsam described the monitors as a near-term step toward a longer research goal: reverse-engineering large language models to trace behavior back to its origins in training.