Frontier safety framework

A developer's published promise of what it will do at each capability threshold.

What it commits you to

Frontier safety frameworks, the Responsible Scaling Policy of 2023 being the first of the genre, pair capability thresholds with pre-committed responses: evaluate, protect, slow. They are the primary governance instrument of the current era and they are voluntary. Their credibility depends entirely on whether the promised response happens when a threshold is actually crossed.

Which realms lean on this

Procedure, not conviction. This is the vocabulary of constitutions, audits and the people who have to sign.

governance · 2 realms — The Contractual Court, The Ubuntu Weave.

Canon that discusses this

declaration · 1986
The Asilomar AI Principles
Drafted over a few hours on the final day of a four-day 1986 conference on future technologies, and the first written set of research principles for advanced AI agreed by the researchers themselves. Its structure — general principles plus separately stated capability thresholds — is the structure of every frontier safety framework since. The excerpt below paraphrases the principles rather than reproducing the agreed wording.

Frontier safety framework

What it commits you to

Which realms lean on this

Canon that discusses this

The Asilomar AI Principles

See also

Capability threshold

Tripwire

Governance brittleness

AI safety case