Technically impossible: who bears the cost of a copyright exception?

Technically impossible." That was Anthropic's special envoy, Jeffrey Bleich, at the parliamentary inquiry on AI this week, describing the prospect of licensing every piece of content needed to train models in Australia.

In the same edition, Capital Brief reported that Australian startup Trellis Data argues the opposite: that Australia can build competitive models without weakening copyright protection, and that its own profitable, homegrown model development shows how.

Both claims can be true, and that is the point policymakers need to grasp. Licensing everything for a frontier general-purpose model may well be impractical. Licensing what is needed for a narrower, commercially useful model may not be. The case for a text and data mining exception rests on the first framing. The case against it rests on the second.

From a legal perspective, I would start with a simpler question: who bears the cost of the exception? A carve-out transfers value from Australian creators to model developers, most of them offshore. That may be a trade worth making, but it should be argued openly as a transfer, not presented as a technical inevitability.

Impossible" usually means "expensive for us". The inquiry should ask how expensive, and for whom.

Source: https://lnkd.in/giDRtJxA

#Copyright #AIPolicy #Australia