The Case of the Leaky AIs
Автор: PPC Land
Загружено: 2026-01-18
Просмотров: 93
Recent research from Stanford University has revealed that prominent artificial intelligence models can be manipulated into reproducing nearly entire copyrighted books from their training data. By employing various prompting techniques, investigators successfully extracted substantial portions of famous novels from systems like Claude, Gemini, and Grok. While some platforms required complex adversarial "jailbreaks" to bypass safety guardrails, others surrendered protected text with minimal effort. These findings highlight significant security vulnerabilities and suggest that current measures to prevent intellectual property theft remain inadequate. This discovery arrives amidst intensifying legal scrutiny regarding how AI developers utilise published works without explicit permission.
Video generated in NotebookLM based on https://ppc.land/stanford-researchers...
Доступные форматы для скачивания:
Скачать видео mp4
-
Информация по загрузке:
