AI study questions authenticity of Van Eyck masterpieces

This story is a small but vivid example of AI systems stepping into domains once ruled entirely by human connoisseurship. A model trained on brushstroke patterns and other visual features is now confident enough to call into question the attribution of works by one of Europe’s canonical painters, with percentages that sound uncomfortably precise. That’s a preview of how AI judgment will increasingly collide with entrenched expert communities—in law, medicine, finance and science, not just art history. ([theguardian.com](https://www.theguardian.com/artanddesign/2026/feb/07/ai-analysis-van-eyck-paintings-turin-philadelphia))

For AGI‑watchers, the significance isn’t that a single model can ‘know better’ than curators, but that hybrid workflows are emerging where AI screens huge spaces of possibilities and humans arbitrate the most consequential calls. If these tools continue to prove accurate, institutions will be under pressure to accept AI‑driven revisions to long‑standing narratives and asset valuations. That same pattern could apply to code audits, safety evaluations, and even scientific peer review as models get better at spotting subtle anomalies.

The controversy also underlines a deeper epistemic risk: over‑trusting seemingly objective probability scores without fully understanding the training data or failure modes. As we move toward more capable systems, building rigorous, transparent validation pipelines—and educating users about what those numbers actually mean—will be as important as raw performance gains.