The role of voluntary safety frameworks is changing
Some companies have introduced separate compliance frameworks. This raises the question of what role their voluntary frameworks should play.
Frontier Risk
Frontier Risk is a newsletter about frontier AI risk management written by the GovAI Risk Management Team
Live Last read · last published · next check
Latest posts
OpenAI’s GPT-5.6
At the request of the US government, OpenAI restricted access to its most capable model. Two weeks later, it was approved for public release.
Anthropic’s Claude Mythos 5 and Fable 5
Anthropic released its most capable model to the public. Three days later, the US government ordered it to restrict access.
Google DeepMind’s Frontier Safety Framework v3.1
Google DeepMind has recently updated its safety framework. It now tracks some risks earlier and commits to a higher security standard for misuse risks.
OpenAI’s GPT-5.5
GPT-5.5’s offensive cyber capabilities are comparable to Mythos. But unlike Anthropic, OpenAI deployed the model publicly.
Comparing Anthropic’s RSP and compliance framework
Anthropic now has two risk management frameworks: its Responsible Scaling Policy (RSP) and Frontier Compliance Framework (FCF). This split has not received enough attention.
Meta’s Advanced AI Scaling Framework v2
Meta recently updated its safety framework. Loss of control is now in scope, and the bar for flagging risky models is lower.
Anthropic’s Claude Mythos Preview
Mythos can autonomously find and exploit zero-day vulnerabilities. That’s why Anthropic restricted access to a few partners. But their RSP would have allowed a public release.
Anthropic’s RSP v3.0
How it works, what’s changed, and some reflections
About the GovAI Risk Management Team
Who we are, what we do, and why we do it