New AI Test Could Make Drug Safety Checks Faster and Safer

Before a new medicine reaches humans, it's tested on lab animals. Pathologists carefully inspect their organs under a microscope to spot damage or disease. This is called toxicologic pathology, and it's a slow, painstaking job. Could AI help? That's the question behind a new study introducing VIPER, the first benchmark designed to evaluate AI models on this kind of veterinary work.

VIPER consists of 1,251 questions paired with 419 microscope images of rat tissue from seven organ systems. Every question was created and checked by board-certified veterinary pathologists, so it's a trusted yardstick. The researchers then put 16 AI models to the test: two brand-new models trained specifically for veterinary pathology, seven trained for human pathology, and seven general-purpose AI systems like the ones powering popular chatbots.

The results are revealing. General-purpose AI models often over-diagnose — they see disease in perfectly normal tissue, which could cause false alarms in drug testing. Veterinary-specific models did better, but there's still a clear gap between what AI can do for human pathology and what it can do for animal pathology. The researchers conclude that domain-specific training is essential, but even then, AI isn't ready to work alone.

Why does this matter to you? The safety of new medicines depends on accurate tissue analysis. If AI can learn to read animal tissues reliably, it could speed up drug development, cut costs, and potentially reduce the number of animals needed in testing. But this study shows we're not there yet — and that's why careful testing like VIPER is a crucial step forward.