AI models risk consuming themselves by depleting human-generated content, creating a self-destructive cycle akin to a mythological curse.
AI models risk self-destruction
Artificial intelligence models may be consuming themselves, according to internal documents from Microsoft and OpenAI. The issue arises as AI systems deplete the human-generated content they rely on for training, creating a dangerous cycle that threatens the models' future development.Ancient myth mirrors modern AI dilemma
In Ovid's Metamorphoses, King Erysichthon is cursed with insatiable hunger after cutting down a sacred tree. He eventually consumes his own body. This myth now seems to parallel the situation of major AI companies, which are consuming vast amounts of human content, risking their own sustainability.Legal battle over AI training data
In late 2023, the New York Times sued OpenAI and Microsoft, alleging unauthorized use of its articles for training AI models. Publishers later joined the lawsuit, and in mid-2026, they submitted a detailed legal brief outlining the damage caused by AI technology to the web.Microsoft admits AI's harmful impact
Microsoft's internal reports describe a "vicious circle" in which its products threaten the economic stability of content creators. The company acknowledges that training AI models on human-generated content represents "the largest theft of human labor in human history."Google's claims contradicted by data
Google previously claimed AI Overviews and AI Mode would not reduce website traffic. However, data show otherwise, with publishers reporting significant drops in traffic. This decline affects how media outlets operate, especially those producing evergreen content.AI's impact on media employment
By late 2026, Prisma Media, a major French publishing group, eliminated nearly 40% of its workforce. The company also trained its journalists to use AI, which now produces up to 40% of its articles, including content from Voici, a leading magazine.AI-generated content harms AI training
AI-generated content is becoming a problem for AI models, as it pollutes the data they use to train. A 2024 Nature study found that models trained on AI-generated data suffer from "model collapse," where they lose the ability to generate realistic content.Tech giants begin compensating content creators
In 2026, Google and Microsoft started initiatives to pay publishers whose content is used by AI. These programs aim to support content creators while ensuring AI models continue to have access to high-quality training data.AI's potential self-sufficiency
Sam Altman of OpenAI once suggested that if AI could generate high-quality synthetic data, they might no longer need human input. This scenario, however, risks creating a self-sustaining system that could lead to unpredictable consequences for both AI and the web.
