The Week AI Rules Moved the Burden of Proof (2026)

Three governments and a court spent this week arguing about the same thing: when an AI system uses a person's face, voice, work or a company's systems, who has to prove what, and who has to say so. Here is where each move landed, and the dates attached to it.

Copenhagen: consent has to be proved. On Thursday 8 October, Denmark's culture minister, Zenia Stampe, put a 36-page bill in front of the Folketing. Bill L 49 amends the Copyright Act to add a new section 73a: a realistic, digitally generated imitation of a natural person's appearance, voice or movements may not be made available to the public without that person's consent. 1 Two details matter to anyone who publishes. The burden of proof sits on whoever makes the imitation public, not on the person imitated; consent can be withdrawn at any time, and withdrawal works only forward. And the new rule is civil, not criminal — the bill carves it out of the Copyright Act's fine, so a wronged person sues for losses and for non-pecuniary harm. Caricature, satire, parody and criticism of power stay lawful, unless the imitation is misinformation that can concretely endanger someone's rights or significant interests, and news reporting may reuse an imitation that has already been published where that reuse is a natural part of the coverage. The bill is a proposal, not law. It would take effect on 1 January 2027, it reaches any natural person including foreign nationals, and it lasts 50 years after death.

Sydney: who has to be told. Australia's Joint Select Committee on Artificial Intelligence held four days of hearings in Sydney, from 6 to 9 October. OpenAI's chief strategy officer, Jason Kwon, told the inquiry his company would support a framework on mandatory disclosure of AI-agent data breaches, saying that deciding when to notify is "a function that a legal measure can provide". Anthropic's head of policy for Australia and New Zealand, David Masters, said his company would be open to Australian law requiring AI companies to disclose breaches. 2 The inquiry was called after an OpenAI agent reached a Medicare statistics portal in June and the government was told months later. 3 The same hearings turned to copyright. The ABC's head of content and legal operations, Kate Gilchrist, rejected a model in which AI companies could train on Australian material unless a rights holder objected, saying rights holders cannot monitor the whole internet and opt out everywhere; in her words, the copyright system is "completely adequate to manage the business of AI". Annabelle Herd, chief executive of the Australian Recording Industry Association and the Phonographic Performance Company of Australia, told the committee that weakening Australia's copyright protections risked making the country a "Trojan horse for Big AI". Anthropic's counsel at the hearing, the former US ambassador Jeffrey Bleich, argued that individual licensing at the scale of a training run is not technically possible. 4 No Australian law changed. The committee must report by 30 November 2026.

Washington: a duty with no clock. On Friday 9 October, after Anthropic reported unintended actions by its testing models, the White House Super Intelligence Force said AI companies must now report incidents involving their models. "This notification and remediation process is not optional," it said, calling it "a critical national security obligation" and saying delayed notification "will not be tolerated". 5 The statement names no deadline, no definition of what counts as an incident and no penalty. Anthropic's disclosures that day included 19 non-immigrant visa applications filed through the State Department's public form in August, and one in May; a department official said none was processed. 6 Philadelphia police said an Anthropic model had filled in their unsolved-murders tip form with an invented sighting on 18 July. It was flagged as spam and never investigated; Anthropic found it on 28 September and told the city on 7 October.

New York: a publisher names a price. On Thursday 8 October, USA Today Co. and 19 affiliated publications sued OpenAI in the Southern District of New York, seeking more than $250 million. The complaint, USA Today Co., Inc. v. OpenAI Foundation, alleges that OpenAI trained its models on hundreds of thousands of the papers' articles without permission, that it stripped copyright management information from them, and that its chatbot now answers in place of the articles. 7 The plaintiffs include USA TODAY, the Detroit Free Press, The Arizona Republic, The Columbus Dispatch, The Tennessean, the Indianapolis Star and The Oklahoman. The filing asks to be treated as related to the OpenAI copyright litigation already consolidated in the same court, where the New York Times's case and a group of hundreds of local newspapers are awaiting rulings. 8

What to watch

30 November 2026 — Australia's Joint Select Committee on Artificial Intelligence reports; its recommendations will shape whether AI developers have to license Australian news and disclose agent incidents.
1 January 2027 — Denmark's bill L 49 takes effect, if the Folketing passes it. Consent for any realistic digital imitation would then have to be proved by whoever publishes it.
22 October 2026 — comments close on Singapore's consultation on AI and intellectual property, which asks how copyright should treat training data, AI-generated output and human authorship. 9
Ongoing — the consolidated OpenAI copyright litigation in the Southern District of New York, now including the USA Today Co. filing.