Robots Atlas>ROBOTS ATLAS

Goodfire

ACTIVE
AI LabStartup
Goodfire AIGoodfire Inc.

An AI research lab specializing in mechanistic interpretability — understanding the internal mechanisms of neural networks so they can be debugged, controlled and intentionally designed. Creator of the Ember platform.

Founded2024United States
HQSan FranciscoUnited States
Total funding$50000B2 · 2025–2026
01 · About

Goodfire

Founded 2024 · San Francisco, United States

Goodfire is an AI interpretability research lab whose mission is to "understand the scientific foundations of neural networks so that we can intentionally design AI." The company works on mechanistic interpretability, studying the internal representations and mechanisms of both large language models (LLMs) and life-science models. Its work is organized around three areas: Understand (reverse-engineering model mechanisms, such as neural geometry and an Alzheimer's biomarker discovered from an epigenetic model), Debug (finding model failures, such as reward-hacking probes and production cyber monitors), and Design (guiding training using internal representations, i.e. "mechanistic shaping"). Its flagship product is the Ember platform, which decodes the neurons inside AI models and gives programmatic access to their internal processes. Goodfire was founded in 2024 by researchers who helped pioneer interpretability at OpenAI and Google DeepMind; its CEO and co-founder is Eric Ho. The company is headquartered in San Francisco and has raised over $200M in total — including a $50M Series A (April 2025, led by Menlo Ventures) and a Series B (February 2026). Its investors include Menlo Ventures, Lightspeed Venture Partners, B Capital and Anthropic. Goodfire has collaborated with the Arc Institute on the Evo 2 DNA foundation model, among others.

  • Eric Ho

    Co-Founder & CEO

    Co-founder and CEO of Goodfire, leading the mission to understand the internal mechanisms of AI models. He publicly argues that because nobody understands the mechanisms by which AI models fail, no one knows how to fix them.

02 · What they do

Domain of activity

2 areas · 4 technologies · 3 industries
What they do2Areas
  • AI developer tools
  • AI Governance & Safety
How they do it4Technologies
  • Mechanistic interpretability
  • Foundation Models
  • LLM
  • AI Security
Who for3Industries
  • Research & science
  • Software & Technology
  • Pharma & biotech
03 · Geography

Global presence

HQ San Francisco · 2 regions
USANorth America
04 · Metrics

Scale & funding

founded 2024 · Series B
2024founded
$50Mtotal funding
Series Blatest round
growthdevelopment stage
Privatenot listed
activestatus
$50M
total funding
2 · 2025–2026
Series AApril 2025Menlo VenturesRound announced in April 2025, less than one year after the company was founded.
$50M
Series BFebruary 2026Series B round announced on 5 February 2026; the specific amount was not stated in the announcement.
amount not publicly disclosed
07 · Classification

Profile & metadata

4 classification · 3 external links

Classification

  • Statusactive
  • Development stagegrowth
  • Listed onNoprivate
  • Organization typeAI Lab+ Startup

External links