A European fertility organization is pushing for strict caps on sperm donor offspring after cases revealed the scale of unregulated reproduction. Ties van der Meer, a 47-year-old conceived through donor sperm, discovered he has an unknown number of siblings from the same donor, highlighting gaps in European oversight.
The fertility group argues that without donation limits, single donors can father hundreds of children across multiple clinics. This creates genetic risks including unintended incest among half-siblings and increases the prevalence of recessive genetic disorders in populations. Countries like Denmark and the UK already enforce caps, typically limiting donors to 10 to 25 families per person.
The lack of standardized rules across Europe means donors can move between clinics and countries to maximize their contributions. Some donors have fathered 100+ children across different facilities, creating a hidden genetic network with minimal transparency. The proposal seeks binding EU-wide limits to prevent this fragmentation.
This issue intersects with AI's "world models" concept, the second part of today's newsletter. World models are neural networks trained to predict how environments evolve over time. They learn patterns from video or simulation data to build internal representations of physics, cause-and-effect relationships, and spatial dynamics. Researchers use these models to pre-train AI systems before task-specific fine-tuning, reducing the computational cost of training from scratch.
OpenAI's recent work on video prediction and Tesla's autonomous driving research both rely on world models. The technology promises to make AI systems more sample-efficient and safer by letting them simulate outcomes before taking real actions. However, world models remain computationally expensive to train at scale and often struggle with long-horizon predictions where small errors compound.
Both stories reflect growing recognition that AI and biotechnology need guardrails. Fertility oversight prevents tangible harms in reproduction. World models require similar caution as they become core infrastructure for