19Open weights & leakageBelow the top 20
Distillation of closed models into open ones
CapabilityDomain knowledgeImpact domainBoth
Statement (NASA form)
Given that APIs can be used as teachers; the student does not need the weights, there is a possibility of a near-frontier open model that no lab released and no lab can recall resulting in safety and access policies that exist only on the teacher.
Likelihood
4Likely
Consequence
4Critical
Urgency
3Priority
If you sell the answers, you sell the student.
Composite 19 = 4×4 + 3
Applicable mitigations
Controls · Capability thresholds that cannot be open-released · Know-your-customer on high-end fine-tune clusters · Pair open release with synthesis and exploit screening upgrades · Hardened weight storage and insider controls · Mandatory vulnerability-sharing with national CERTs before launch · Staged release with external red-team gates