AI Safety & Society · The 2026 incident files
You can name the three landmark AI copyright actions of 2025-26 and describe what each one is actually deciding.
For two years, "AI copyright" was a single argument with no clear winner. By mid-2026, three specific case files had done most of the work of separating the questions.
Bartz v. Anthropic split the acquisition question from the training question. Judge Alsup held that the training use and print-to-digital conversion of purchased books were fair use, while the pirated central-library copies were not excused by fair use on the summary-judgment record. The official settlement site then became the operational file for resolving the piracy side.
Getty Images v. Stability AI went the other direction in the UK High Court in November 2025: the live secondary-copyright claim failed because the court did not treat Stable Diffusion weights as copies of Getty works, while limited trademark findings survived.
NYT v. OpenAI/Microsoft sits between them. The case is still in motion, and its discovery orders turn the abstract training-data question into concrete fights over output logs, preservation, inspection, and what a plaintiff can force a model developer to retain.
This chapter walks each case and then asks what these decisions, plus the 10^26 FLOPs threshold from frontier-AI statutes, mean for the next wave of suits.
Type: multi-choice
Prompt: > Why does it matter that Bartz, Getty, and NYT are being decided separately rather than as one consolidated case?
Chapter contains 4 lessons.