Anthropic Proposes Opt-Out AI Training Copyright in Australia

Anthropic has urged the Australian government to implement an opt-out copyright licensing scheme for artificial intelligence training data. This proposal, submitted to the Attorney-General’s department, seeks to permit AI models to ingest copyrighted material unless the rights holder explicitly forbids it. The firm suggests a system requiring rights holders to register their refusal for content to be used in AI training.

The submission contends that a broad opt-in regime for AI training is impractical. Requiring individual licensing agreements for every piece of copyrighted work would be excessively burdensome. Such a system could stifle innovation within Australia's burgeoning AI sector. Conversely, an opt-out mechanism could facilitate the training of larger, more capable AI models. This approach might also help Australian AI companies compete internationally.

Anthropic’s proposal references the American legal principle of 'fair use.' This doctrine allows for the unlicensed use of copyrighted material under specific circumstances, often for transformative purposes like commentary or parody. The company argues that training AI models constitutes a similarly transformative use. It suggests that AI models learn patterns from data, rather than reproducing the original works.

Several jurisdictions currently grapple with how to adapt copyright law for AI. The UK government previously considered a similar opt-out exception but reversed its decision in early 2023. This reversal followed significant pushback from creative industries, who argued such a change would devalue their work. Australia's existing copyright framework, similar to the UK's, lacks a broad fair use provision, instead relying on specific 'fair dealing' exceptions.

The Attorney-General's department initiated its inquiry in July 2023, seeking to update Australia's copyright laws for the digital era. The department’s consultation paper acknowledged that AI training could infringe copyright if it involves reproducing material without permission. Anthropic's intervention represents a significant voice in this ongoing debate. The outcome of Australia's review could set a precedent, influencing how other nations approach AI and intellectual property. Will global copyright norms shift to accommodate AI’s data demands?