Goodfire and Baseten partner to launch Project Beacon safety monitoring
Start here
Goodfire and Baseten are collaborating on Project Beacon, integrating Goodfire's safety stack into the Baseten inference platform, defining policies by reading the model's internal activations during generation, detecting and handling unsafe events before outputs reach users or tools, covering four risk categories: prompt injection, unauthorized operations, sensitive data leakage, and network abuse. Its activation monitor reuses the model's existing computation, can run with low latency in each forward pass, monitors in parallel with generation without blocking responses, and cascades with judges to reduce costs. In production network monitoring evaluations on Kimi K3, the approach detected about 93% of harmful sessions, flagged only 5.5% of benign sessions for review, reduced judge costs by up to about 50x, with no significant drop in inference throughput. Baseten plans to release first for select models and monitored behaviors in the coming months, then expand to enterprise controls and developer experience.
Generated by AI from reports · Updated 3 days ago
Progress timeline
- BasetenBaseten and Goodfire AI launch Project Beacon, bringing inline safety controls to open model inference
Baseten announced a partnership with Goodfire AI to launch Project Beacon, introducing proactive, inline safety controls at the model inference layer, defining policies by reading the model's internal activations during generation, and detecting and disposing of unsafe events before outputs reach users or tools. The approach targets four risk categories—prompt injection, unauthorized actions, sensitive data leakage, and network abuse—combining activation probes with text monitoring, with monitoring running in parallel with generation without blocking response generation. Baseten plans to first release for select models and monitored behaviors in the coming months, then expand to enterprise controls and developer experience, and bring it to market with a small set of early partners.
- Goodfire ResearchGoodfire and Baseten partner to bring frontier safety monitoring to open-source models
Goodfire and Baseten partner to launch Project Beacon, integrating Goodfire's safety stack into the Baseten inference platform to detect unsafe behavior in open-source model agents in real time. Its activation monitors reuse the model's existing computation, run at low latency in each forward pass, and cascade with a judge to reduce cost. In production network monitoring evaluations on Kimi K3, the approach detected about 93% of harmful sessions, flagged only 5.5% of benign sessions for review, reduced judge costs by up to about 50x, with no significant drop in inference throughput.
Related discussion
Attention trend
New sources per day
- Oct 10New2 sources (2 outlets,0 accounts)
- Oct 11New0 sources (0 outlets,0 accounts)
- Oct 12New0 sources (0 outlets,0 accounts)
- Oct 13New0 sources (0 outlets,0 accounts)