Ex-OpenAI DeepMinders bag $150M for tools that debug AI hallucination
2 min read
AI Funding News
February 6, 2026
Most current AI models operate as black boxes, capable of writing, predicting, and reasoning without clear insight into why they produce specific outcomes. This lack of transparency complicates control, troubleshooting, and safe deployment.
San Francisco-based AI research lab Goodfire aims to solve this problem, securing $150 million in a Series B round at a $1.25 billion valuation. The funding was led by B Capital, with participation from Menlo Ventures, Lightspeed Venture Partners, South Park Commons, Wing Venture Capital, DFJ Growth, Salesforce Ventures, and Eric Schmidt.
The capital will support Goodfire’s development of a “model design environment”—a platform enabling developers to understand, debug, and intentionally design AI systems at scale. The company will also advance its research into fundamental model interpretability methods.
Making AI Systems Transparent
Led by CEO Eric Ho, Goodfire focuses on building AI that is both powerful and interpretable. The team includes alumni from OpenAI, DeepMind, Stanford, and Harvard, backed by over $200 million in total funding.
Yan-David “Yanda” Erlich, General Partner at B Capital, emphasized the need for deeper model understanding: “Bridging the gap between tracking AI behavior and truly comprehending why models act as they do is critical. Goodfire’s work unlocks safer, more useful AI by enabling intentional design.”
Inside Goodfire’s Approach
Unlike traditional retraining methods, Goodfire’s technology targets specific internal components within models to adjust behavior directly. For example, the company reduced hallucinations in a large language model by nearly 50% through precise internal tweaks. Its interpretability tools have also advanced scientific research, such as identifying new Alzheimer’s biomarkers in collaboration with the Mayo Clinic and Arc Institute.
Goodfire represents a new wave of AI “neolabs”—research-driven organizations pursuing breakthroughs in model understanding that are often overlooked by larger scaling-focused entities like OpenAI or Google DeepMind.
“Interpretability is the foundation for designing intelligence intentionally, not stumbling into it,” said Ho. “Every engineering discipline has relied on fundamental science, and AI is now at that pivotal moment.”
The team includes Nick Cammarata, a core contributor to OpenAI’s interpretability research; Tom McGrath, founder of DeepMind’s interpretability team; and UC San Diego professor Leon Bergen (on leave).
About Author
