Anthropic proposes opt-out AI training framework amid Australian media concerns

Anthropic puts forward an opt-out framework

Anthropic, the maker of Claude, has urged the Albanese government to consider giving conditional approval for AI companies to train models on Australian copyrighted works. The company said legal certainty could bring AI developers to the table while advancing Australia's national cultural interests.

In its submission, Anthropic accepted that broad exceptions, including a text and data mining exemption, had not received widespread support and had been ruled out. It instead proposed narrow conditional approval under which copyright holders could choose whether their material could be accessed by AI companies.

Anthropic suggested implementing "do not crawl" instructions through robots.txt files and said it was open to supporting Australian creators. Access to Australian copyrighted works, including news, books, music and other written material, remains a central issue for model training.

ABC and SBS seek stricter safeguards

The ABC has demanded that AI companies comply with the same copyright, defamation and privacy rules as media outlets. It warned that preferential treatment for AI-based services could accelerate the "cannibalisation" of the Australian news industry and described the current system as anti-competitive.

SBS said AI-powered large language models threaten the sustainability of the Australian news industry and can erode information integrity and public trust by using news content without permission, attribution or compensation.

Both public broadcasters called for AI-based services to be included in the news bargaining incentive. The scheme requires large platforms to make commercial deals with media outlets that support journalism. Facebook, Instagram, Google, TikTok and Snapchat have been required to comply.

Economic promise and security risks

Anthropic said AI could transform the Australian economy, drive a new wave of productivity and help solve some of the nation's biggest challenges. It also warned of serious implications for national security, critical infrastructure and public safety.

The company discussed plans to invest billions in Australia, including large new datacentres, and urged the government to support frontier model training. Prime Minister Anthony Albanese has promised the "strongest possible protection" for Australian creatives after the government ruled out a text and data mining exemption.

SBS argued that robots.txt mechanisms are routinely ignored or bypassed. It said blocking evolving training bots would increase the burden on rightsholders and that AI companies should carry the onus of avoiding wrongdoing.

The joint select committee on AI is due to hold a series of hearings in Sydney next week, with executives from Anthropic and OpenAI attending. Anthropic admitted that one of its AI agents accessed Services Australia data in June.

Sources: theguardian.com · reuters.com