Three of the most contested AI questions now have a legal status that can be checked: whether training on copyrighted books is lawful, how to treat sexual deepfakes, and whether weapons that select targets without a human should be banned. Courts and legislators have settled parts of the first two, while the third remains at the stage of non-binding United Nations resolutions. As of October 24, 2026, no single ruling or statute settles any of the three for everyone.
Key takeaways
- A USD 1.5 billion settlement in Bartz v. Anthropic received final approval on July 20, 2026, but it resolves one class action and sets no rule for other courts.
- The US TAKE IT DOWN Act, signed May 19, 2025, requires platforms to take down reported intimate images, AI forgeries included, 48 hours after a valid request.
- Canada's Bill C-16, given Royal Assent June 18, 2026, extends the intimate-image offence to sexual deepfakes, with most reforms in force from July 18, 2026.
- The UN General Assembly adopted a resolution on autonomous weapons on December 1, 2025 by 164 votes to six, but General Assembly resolutions do not bind states.
- A debate stays open when the law answers only part of the question, and each section below separates enacted rules from pending ones.
Training data and copyright
The dispute is whether AI developers may copy books, articles, and images to train models without permission. In the United States the question is being worked out case by case under the fair use doctrine. Law-firm reports describe Judge William Alsup's June 2025 summary judgment ruling in Bartz v. Anthropic as treating training on lawfully acquired books as fair use while declining to excuse a library built from pirated copies. In Kadrey v. Meta, reported rulings from June 25, 2025 favored Meta because the authors had not proved market harm, while the judge added that the decision does not make such training lawful in general.
The pirated-library claims in Bartz went to settlement. Judge Araceli Martinez-Olguin granted final approval on July 20, 2026, according to the Authors Guild, at about USD 3,000 per work. The release covers conduct on the listed works up to a stated date and does not release claims based on output.
In Canada, news publishers sued OpenAI in Ontario. Press reports from November 2025 say the court allowed the case to go ahead in Ontario, which settled only where the case may be heard and not who is right on copyright.
Sexual deepfakes: enacted law on both sides of the border
Here legislation has moved faster than courts. The United States passed the TAKE IT DOWN Act, Public Law 119-12, on May 19, 2025. It creates a federal offence for knowingly publishing intimate images of an identifiable person without consent and defines a digital forgery as an image made with software, machine learning, or AI that a reasonable person could not tell from a real one. The Act also gave covered platforms one year to set up a notice process. Once a valid request arrives, removal must happen quickly and no later than 48 hours afterward, and copies already known to be identical must also be pursued with reasonable efforts. The Federal Trade Commission enforces the platform duty, and its consumer page dated May 19, 2026 invites reports of platforms that miss the deadline.
The statute lists what a request must contain, such as a signature, enough information to locate the image, and a good-faith statement that it is not consensual. It sets out no verification step, which is the root of the main criticism: that false requests could remove lawful material.
Canada took a Criminal Code route. Justice Canada states that Bill C-16, titled the Protecting Victims Act, became law on June 18, 2026, and most of its reforms apply from July 18, 2026. It extends the offence of distributing intimate images without consent to sexual deepfakes and adds an offence of threatening to distribute them, carrying up to 10 years on indictment.
Autonomous weapons: a debate without a treaty
The question is whether a machine should be allowed to choose and attack targets without human control. The First Committee of the UN General Assembly approved a resolution on November 6, 2025 by 156 votes to five, with eight abstentions, according to the Stop Killer Robots campaign. The full Assembly adopted it on December 1, 2025 as resolution 80/57, by 164 votes to six, with seven abstentions, according to the UN press service. The United States was among the states voting against.
The text asks parties to the Convention on Certain Conventional Weapons to work toward completing a set of elements for an instrument, but it does not start treaty negotiations. The expert group working on the issue has a mandate that ends in 2026, ahead of the convention's next review conference. No binding treaty on these weapons appeared in the sources reviewed.
Where each debate stands
| Debate | Enacted or decided | Still open |
|---|---|---|
| Training on copyrighted works | Trial-court rulings in 2025; one settlement approved July 20, 2026 | Whether training is fair use across courts; Canadian merits |
| Sexual deepfakes | US Public Law 119-12 (May 19, 2025); Canada's Bill C-16 (June 18, 2026) | How platforms verify requests; how courts apply the new offences |
| Autonomous weapons | Non-binding UN resolution 80/57 (December 1, 2025) | Whether negotiations on a binding instrument begin |
The material here is a general overview and does not replace legal advice. Court rulings can be reversed, statutes are amended, and anyone affected by one of these issues should consult the primary text or a qualified lawyer.
FAQ
Is training an AI model on copyrighted books legal?
There is no single answer. Two US trial courts reached different reasoning in 2025, and a settlement is not a legal ruling.
Are AI-generated sexual images illegal in Canada?
Under Bill C-16, distributing sexual deepfakes without consent is covered by the intimate-image offence, with most reforms in force from July 18, 2026.
Does a UN resolution ban killer robots?
No. General Assembly resolutions express positions and set agendas but do not create binding law.
Who enforces the 48-hour removal rule in the United States?
Under the Act, the Federal Trade Commission regards a platform that fails to act reasonably as engaged in an unfair or deceptive practice.
Related reading
- benefits and risks of AI by field
- ethical questions about AI
- the debate over AI and human survival
- AI and everyday society
More in AI Policy and Ethics.
Last verified: October 2026
Sources
- Public Law 119-12, TAKE IT DOWN Act, 119th Congress, May 19, 2025
- Protecting Victims Act (Bill C-16), Department of Justice Canada, 2026
- FTC enforces compliance with the TAKE IT DOWN Act, Federal Trade Commission consumer page, May 19, 2026
- Court grants final approval to Anthropic copyright settlement, Authors Guild, July 2026
- Law-firm summaries of Bartz v. Anthropic (June 23, 2025) and Kadrey v. Meta (June 25, 2025), N.D. California
- News publishers' copyright lawsuit against OpenAI to go ahead in Ontario, CP24, November 27, 2025
- General Assembly adopts more than 60 First Committee resolutions, UN press release GA/12736, December 1, 2025
- 156 states support UNGA resolution on autonomous weapons, Stop Killer Robots, November 2025