Australia: Anthropic offers an opt-out option for training its models on copyrighted works; the ABC condemns the ‘cannibalisation’ of information

In the absence of a general exception to copyright, Anthropic suggests a ‘conditional licence’; the ABC and SBS, on the other hand, are calling for AI to be subject to media regulations.

In a contribution to the Australian Parliamentary Inquiry into AI, Anthropic takes note of the Albanese government's refusal to introduce a text and data mining exception. Instead, the company is proposing a restricted "conditional authorization": training would be allowed as long as rights holders have not signaled their refusal, for example via the robots.txt file. It highlights its plans to invest billions in Australia, including large data centers, while acknowledging risks to national security and critical infrastructure.

Public broadcasters stand up

The ABC and SBS are challenging this approach head-on. According to the ABC, robots.txt guidelines are routinely ignored or circumvented, and it is up to AI companies, not publishers, to prove that they respect rights. The broadcaster considers it "anticompetitive" that AI services escape copyright, defamation and privacy obligations on the media, at the risk of accelerating the "cannibalization" of the Australian press.

The two groups are calling for AI companies to be included in the mandatory negotiation mechanism with the media, already imposed on Google, Meta, TikTok and Snapchat. Executives from Anthropic and OpenAI are due to be heard next week in Sydney by the committee.