01 / The short version
What happened
Meta updated its Superintelligence Scaling Framework, setting standards for training, evaluating, and releasing capable AI models. The update extends loss-of-control requirements to cover AI containment during training and evaluation, requiring mitigations before reinforcement learning runs if a model could compromise its environment. It also addresses open-weight releases and governance, including a new Board AI committee to review Framework changes and ensure compliance with commitments made at the White House.
See the exact references02 / Key takeaways
What you need to know
- 01
Meta extends loss-of-control requirements to training and evaluation containment.
- 02
Framework now covers open-weight release considerations.
- 03
A new Board AI committee will review Framework changes and compliance.
Keep in perspective
What to watch for
The source does not specify which models or release dates the updated thresholds apply to.
Go to the source
Exact references
These are the original pages used for this brief. Publisher claims are not independent evaluations.
01Primary source · MetaRead the original announcementhttps://research.meta.ai/blog/developing-capable-models-responsiblyAI-generated from the linked source. It can miss context; verify consequential details in the original. How the radar works