Organisation
Goodfire
A mechanistic-interpretability startup building tools to inspect, debug and steer the internals of neural networks, sold as a commercial platform.
Goodfire is a San Francisco startup, founded in 2024 by Eric Ho, Daniel Balsam and Tom McGrath, that works on mechanistic interpretability — reverse-engineering the internal computations of neural networks so that model behaviour can be inspected, debugged and steered rather than treated as a black box. It develops a commercial interpretability platform and applies the techniques across language models and scientific and biological models. In April 2025 it raised a $50m Series A led by Menlo Ventures that included Anthropic, reported as Anthropic's first investment in an outside startup. It is one of the more heavily funded companies trying to turn interpretability research into a product.
- Category
- Safety & alignment research
- Founded
- 2024
- HQ
- San Francisco, US
- Funding
- $50m Series A (2025), led by Menlo Ventures, with Anthropic
- Key people
- Eric Ho, Tom McGrath