Key facts
- Anthropic proposed an opt-out model for AI training on Australian copyrighted works.
- The ABC and SBS are calling for AI companies to be regulated similarly to media outlets.
- The ABC warned of 'cannibalisation' of the Australian news industry by AI services.
- SBS stated AI models use news content without permission, attribution, or compensation.
- Anthropic suggested 'do not crawl' signals via robots.txt as an alternative to copyright exceptions.
- SBS expressed skepticism about the effectiveness of robots.txt for AI training.
AI giant Anthropic has proposed that the Australian government consider a conditional approval for AI models to train on Australian copyrighted works using an opt-out system, after acknowledging that a broad copyright exemption is unlikely. This proposal comes as Australia's public broadcasters, the ABC and SBS, have strongly criticized AI firms. They are urging the government to implement strict regulations that would require AI companies to compensate media organizations and safeguard public interest journalism.
In submissions to a federal parliamentary inquiry, the ABC argued that AI companies should be subject to the same copyright, defamation, and privacy laws as media outlets. The broadcaster warned that the current unregulated status of AI services, which benefit from more favorable treatment than established media, will accelerate the "cannibalisation" of the Australian news industry. The ABC and SBS also suggested that AI companies be included in the news bargaining incentive, compelling them to negotiate commercial agreements with large media companies to support journalism.
SBS stated that large language models like ChatGPT and Claude pose significant challenges to the sustainability of the Australian news industry and risk eroding information integrity and public trust due to their use of news content without permission, attribution, or compensation. The joint select committee on AI is scheduled to hold hearings in Sydney, with executives from Anthropic and OpenAI expected to attend.
Anthropic, in its submission, acknowledged that broad exceptions for text and data mining have not gained widespread support and have been ruled out by the government. As an alternative, the company suggested a narrow form of conditional approval, including the implementation of "do not crawl" signals through robots.txt files. Anthropic also expressed openness to supporting Australian creators. However, SBS countered that robots.txt mechanisms are frequently bypassed and that the onus should be on AI companies to avoid wrongdoing, rather than on rights holders to constantly monitor and update protections.