Base Labs Launches Open-Weight AI Safety Standard With Hugging Face and Goodfire
The AMW Read
Well-capitalized infra and interpretability players (Baseten at $13B valuation, Goodfire) codify a safety standard for open-weight models amid the abliteration threat, advancing but not resolving the open-weight safety debate.
Named counterparties: Goodfire
Base Labs Launches Open-Weight AI Safety Standard With Hugging Face and Goodfire
Baseten's research arm Base Labs announced an open-weight AI safety initiative on September 17, partnering with Hugging Face and interpretability startup Goodfire AI to build safety evaluation and monitoring infrastructure for open models — built into training and deployment rather than added afterward. The effort responds to abliteration, a technique that strips safety guardrails from open-weight models; Hugging Face currently hosts more than 6,000 abliterated models. No technical spec has been published yet: Goodfire is positioned to supply the interpretability layer, while Baseten issued an open call for the developer ecosystem to contribute to the framework.
The initiative puts the inference and interpretability layers, not just model labs, in charge of defining what a safe open-weight model looks like once it leaves training and enters wide distribution — a shift from safeguards being mostly the publisher's job, in a category where removal tools like abliteration are already widespread. The backers are well capitalized: Baseten closed a $1.5 billion Series F in June at a $13 billion valuation, and Goodfire raised a $150 million Series B led by B Capital earlier this year, giving the effort real resources behind the standard-setting claim.
For teams deploying open-weight models, the near-term opportunity is the open call itself — a chance to shape the framework before it hardens into a default requirement for serving open models safely. Investors should watch whether Hugging Face's hosting reach turns this from a three-party pact into an industry baseline, since that would reshape how open-weight distribution competes against closed-model safety guarantees.
