AI Scraping and Copyright: What the Government's Three Options Would Mean
The Albanese government is considering copyright changes that would give AI companies a legal right to scrape online content for training, provided they sign enough high-level deals with groups that speak for authors, musicians and other copyright holders.
The detail is still fluid. As of 17 September 2026 the proposals had not been finalised and the government was still consulting, after a leaked government document was tabled in the Senate on Tuesday setting out the options under examination, as detailed by the Guardian.
Three models are on the table. The first would let AI developers access and train models on any online material that is not digitally protected, as long as they have made deals with enough businesses. Text and data mining, the automated copying of large amounts of material to train a model, would be allowed once that deal threshold is met.
The second model would work through rights-holder organisations. An AI company would make a broad deal covering a category of content, such as music or text, that would apply to many small creators in that industry. The organisation negotiates the category deal. The individual creator is covered by it, whether they sat at the table or not. Copyright students will know this shape as extended collective licensing.
The third model would require AI companies to pay a central body that does not yet exist. That body would then distribute payments to "registered" rights holders. Registration, rates, distribution rules and governance would all need to be built.
For individuals, the default would change. A person who did not want their work scraped would need to protect it digitally, such as with a password or paywall. Material placed on the open internet without that protection would be available for training. Material posted on social media would be different again. The platform, not the individual user, would control AI scraping access.
Rights-holder groups have raised two objections to the opt-out. They say not all scrapers observe the robots.txt protocol, the small file that asks crawlers not to collect a site. They also say using robots.txt to opt out can remove content from search results. Content hidden from the crawler is also hidden from Google.
There is an international element. Overseas creators would have the option to block Australian access to their music and videos if AI companies were granted sweeping access to content in Australia. That would mean geo-blocking, limiting access by location, as a form of copyright self-defence.
In July, the government said it had no plans to grant a text and data mining exemption allowing AI companies to scrape content to train models, as reported by the Guardian. The leaked options set out a broad mining right with deal-making and compensation mechanisms attached.
Other July material records further background. A proposal put to cabinet would have allowed AI companies to mine content in exchange for investment and a $350m fund to compensate artists, according to earlier reporting in the Guardian. An industry push sought a copyright carve-out for AI companies in exchange for at least $50bn in datacentre investment. Senator David Pocock's office was tipped off about that industry push in late June. In the same month, the government said it will establish Australian Standards for AI that are clear, consistent and mandatory, according to the Prime Minister's office.
Senator David Pocock criticised the idea, saying Labor easing copyright protections for AI datacentre investment would throw creatives and copyright holders "under the bus", as reported by the Guardian. Flinders University lecturer Lisa Harrison said tech companies likely already scrape online content to train AI models and the proposed reforms could give them the legal right to do so. The government is separately described as considering reforms that would give US big tech unrestricted access to Australian content for AI scraping. In the United Kingdom, the government rejected similar AI copyright reforms after opposition from artists including Elton John and Paul McCartney.
The broader context here is worth spelling out for people who follow policy. Current law starts with permission. These options start with access, then add bulk deals, a future collecting body or a technological lock as the backstop. Creators opt out by paywalling work. Small publishers opt out by disappearing from search. Platforms decide for their users. Overseas owners decide by withdrawing from Australia. When Elton John and Paul McCartney agree in public — Whitehall listens, and Canberra should too.
Looking at what comes next, the consultation matters more than the leak. A central distributor only works if registration is simple, payments are transparent and AI firms cannot game the definition of enough deals. A category deal only works if the organisation genuinely speaks for the long tail it claims to cover. And a protect-it-yourself rule only works if voters accept that posting a photo, a poem or a council submission online means feeding a model unless they lock it away. That is a big ask. Expect the crossbench to press on the July denial, on the size of the investment sweetener, and on whether Australian standards that are clear, consistent and mandatory can coexist with an open scraping right. The government says it is still consulting. It will need to show its working.
