As automated attack techniques rapidly advance, CAPTCHAs remain a critical defense mechanism against malicious bots. However, existing CAPTCHA schemes encompass a diverse range of modalities—from static distorted text and obfuscated images to interactive clicks, sliding puzzles, and logic-based questions—yet the community still lacks a unified, large-scale, multimodal benchmark to rigorously evaluate their security robustness. To address this gap, we introduce MCA-Bench, a comprehensive and reproducible benchmarking suite that integrates heterogeneous CAPTCHA types into a single evaluation protocol. Leveraging a shared vision–language model backbone, we fine-tune specialized cracking agents for each CAPTCHA category, enabling consistent, cross-modal assessments. Extensive experiments reveal that MCA-Bench effectively maps the vulnerability spectrum of modern CAPTCHA designs under varied attack settings, and—crucially—offers the first quantitative analysis of how challenge complexity, interaction depth, and model solvability interrelate. Based on these findings, we propose three actionable design principles and identify key open challenges, laying the groundwork for systematic CAPTCHA hardening, fair benchmarking, and broader community collaboration.

Medical semantic-mask synthesis enhances data augmentation and analysis, yet most GAN-based methods still generate one-to-one images without maintaining spatial consistency across slices. To overcome this, we propose AnatoMaskGAN, a novel framework that embeds interslice spatial features, integrates diverse augmentation strategies, and optimizes deep feature learning for complex medical scans. Specifically, a GNN-based slice-feature fusion module models spatial relationships and aggregates contextual cues from neighboring slices, capturing anatomical structures more comprehensively. A 3D spatial noise-injection strategy fuses weighted spatial features with noise to enrich structural diversity, while a grayscale-texture classifier refines grayscale and texture representation during generation. Experiments on L2R-OASIS and L2R-Abdomen CT show that AnatoMaskGAN achieves 26.50 dB PSNR and 0.9229 SSIM on L2R-OASIS, surpassing the state of the art by 0.43 dB, and reaches 0.8602 SSIM on L2R-Abdomen CT, a 0.48-point improvement. Ablation studies confirm that removing any of the three modules degrades PSNR, SSIM, and LPIPS, verifying the effectiveness of each component in enhancing reconstruction fidelity and perceptual quality. Code is available at https://github.com/noheadwuzonglin/AnatoMaskGAN.