Two stories about artists who cannot speak for themselves arrived this week from different galaxies of the cultural discourse. Hilma's Ghost, a feminist art collective, is reclaiming Hilma af Klint's legacy and the esoteric spiritual traditions her work drew from, traditions that institutional art history spent decades minimizing. Simultaneously, a 2026 arXiv paper by Grimes and Harrison presents BLK-Assist, a modular framework for artist-specific fine-tuning of diffusion models, essentially a methodology for training an AI on a living or dead artist's style to generate new work in their voice. Both are acts of speaking for an artist. One is explicitly feminist and recuperative. The other is technically neutral and therefore ethically unexamined.

Who Reclaims and Who Replicates

The distinction between Hilma's Ghost and BLK-Assist is partly intent and partly consent. Hilma af Klint died in 1944, instructing that her abstract paintings be kept hidden for 20 years after her death. Her estate eventually released them. A feminist collective reinterpreting her spiritual practice is an act of historical argument. An AI framework trained on her visual style to generate new images is something closer to extraction. But the law does not yet distinguish between them with any precision, and the art market certainly does not. Larissa Pham's new novel on predatory mentorship in the art world is a fictional account of a similar dynamic: powerful figures using less powerful artists' labor, affect, and voice as raw material for their own authority.

The Authorship Stack in 2026

What these stories collectively describe is an authorship stack that has never been more complex or more contested. At the bottom: the original artist, living or dead, with or without consent mechanisms. In the middle: the interpreter, whether collective, institution, or AI framework. At the top: the market and the platform, which determine whose interpretation gets amplified and whose gets erased. for funds that have articulated artist-rights positions, because LP pressure on this issue is growing faster than most technical teams expect. The ghost in the machine is increasingly a real person's ghost. That requires a more careful haunting.