MIT's New Method Flags AI Models Trained on CASM Without Generating It

MIT researchers develop a method called Gaussian probing to identify AI models fine-tuned to generate child abuse imagery without generating images, achieving 100% accuracy.