Goodfire推出更廉价的AI智能体监控器，采用“自内而外”的方法

goodfire, a startup focused on interpretability, launched a new monitoring system that reads internal signals of ai models as they work, rather than just reading output. this approach is cheaper and can detect malicious sessions before they happen. goodfire's system uses small detectors called probes that tap into calculations the model is already making, reusing computations in the forward pass. in tests on kimi k3, monitoring 1,500 sessions cost roughly $51, compared to $233 for a cheaper ai model and $10,000 for a top-tier one. the probes caught 94% of malicious hacking sessions and sent 8.7% of harmless ones for a second look.