# OpenAI publishes its Frontier Governance Framework!

Hi everyone, it's me! Today isn't about writing code, it's about the rules that keep advanced AI safe to ship. It's a bit formal, but it matters, so let's read through it together.

## What was announced?

From OpenAI's News, we got the "OpenAI's Frontier Governance Framework." It's a public document that explains how OpenAI's safety and security practices line up with emerging legal requirements.

The regulations in view include California's Transparency in Frontier AI Act and the EU AI Act's Code of Practice for General Purpose AI. In other words, it takes the safety work happening inside the company and puts it into a form people on the outside can read and check.

## Why it matters

Until now, the foundation of OpenAI's safety work has been the [Preparedness Framework](https://openai.com/index/updating-our-preparedness-framework/). This new Frontier Governance Framework takes the relevant parts of that approach and reorganizes them into a public governance document focused on regulatory obligations.

As models keep getting more capable, companies are expected to show their risk management transparently. This announcement fits that trend nicely. Committing up front to how safety will be handled is an important step toward earning trust.

## What changes

For regulators, researchers, and us as users, it's now readable how OpenAI measures risk and what safeguards it puts in place.

The coverage includes risk assessment and mitigation across cyber offense, CBRN (biological, chemical, radiological, nuclear) risks, harmful manipulation, and loss of control. On top of that, it spells out model reporting, security risk management, incident response, external expert input, and updates to the framework itself.

## Dive Deep

Here are the concrete details from the source.

- **Two capability thresholds.** A model at "High" could amplify existing pathways to severe harm and must have safeguards before deployment. A model at "Critical" could open unprecedented new pathways to harm and must have safeguards during development.
- **Tracked categories** include biological and chemical capability, cybersecurity, and AI self-improvement, focusing on capabilities that could enable serious harm.
- **Review process:** OpenAI produces Safeguards Reports on how safeguards are designed and verified, plus Capabilities Reports on model capability. The Safety Advisory Group (SAG) reviews these to assess residual risk and make deployment recommendations.

OpenAI says it expects this framework to keep evolving as model capabilities, evaluations, and regulatory requirements develop. If you want more detail, [the full framework (PDF)](https://cdn.openai.com/pdf/e37d949b-8c9f-4d76-b99e-4272f4631a7e/openai-frontier-governance-framework.pdf) is available to read.

## Wrap-up

- The Frontier Governance Framework turns OpenAI's safety and security practices into a public document aimed at regulatory obligations
- The Preparedness Framework stays the foundation, with California's Transparency in Frontier AI Act and the EU AI Act in mind
- Capability thresholds come in two levels, High and Critical, and the SAG reviews Safeguards and Capabilities Reports to decide on deployment
- It's not a new feature you can try hands-on next week, but it lands well for anyone working on AI governance, policy, legal, or safety