Goodfire, a San Francisco-based AI interpretability research company, has secured $50 million in Series A funding led by Menlo Ventures. The investment, which also includes participation from Lightspeed Venture Partners, Anthropic, and B Capital, will support the expansion of Goodfire’s research initiatives and the development of its flagship interpretability platform, Ember.
The funding comes less than a year after Goodfire’s founding, with its team comprising top AI interpretability researchers and experienced startup operators from organizations like OpenAI and Google DeepMind. Goodfire’s researchers have helped establish the field of mechanistic interpretability, authoring key papers and pioneering advancements such as Sparse Autoencoders for feature discovery.
Goodfire’s Ember platform decodes the neurons inside AI models, providing direct access to their internal workings. This allows users to discover hidden knowledge, shape model behaviors, and improve performance. The company’s technology aims to address the knowledge gap surrounding neural networks, making them easier to engineer and less prone to unpredictable failures.
The investment reflects the growing importance of mechanistic interpretability in AI development. “As AI capabilities advance, our ability to understand these systems must keep pace,” said Dario Amodei, CEO and Co-Founder of Anthropic. Goodfire plans to accelerate its research through targeted initiatives with frontier model developers and release additional research previews across diverse fields.
Goodfire has already collaborated with Arc Institute, with Patrick Hsu, co-founder of Arc Institute, stating that Goodfire’s interpretability tools have enabled the extraction of novel biological concepts from their DNA foundation model, Evo 2. The company’s efforts aim to reveal new scientific insights and reshape the understanding of AI models.
