{"data":{"id":"ca5c5947-e54f-42f4-a35e-734813d13dbf","title":"Generalizable Synthetic Image Detection via Language-Guided Contrastive Learning","summary":"AI can now create extremely realistic fake images using generative adversarial networks (GANs, which generate images by having two competing neural networks work against each other) and diffusion models (AI systems that create images by gradually removing noise). While this technology has legitimate uses, it poses serious risks like spreading misinformation and creating fake profiles, and existing detection methods struggle to identify images from new, unseen generation models. This research proposes a detection method using language-guided contrastive learning (a technique where an AI learns to distinguish real from fake images by comparing them against text descriptions, helping it recognize synthetic images it hasn't encountered before).","solution":"N/A -- no mitigation discussed in source.","labels":["research","safety"],"sourceUrl":"http://ieeexplore.ieee.org/document/11281880","publishedAt":"2025-12-08T13:17:27.000Z","cveId":null,"cweIds":null,"cvssScore":null,"cvssSeverity":null,"severity":"info","attackType":[],"issueType":"research","affectedPackages":null,"affectedVendors":[],"affectedVendorsRaw":[],"classifierModel":"claude-haiku-4-5-20251001","classifierPromptVersion":"v3","cvssVector":null,"attackVector":null,"attackComplexity":null,"privilegesRequired":null,"userInteraction":null,"exploitMaturity":null,"epssScore":null,"patchAvailable":null,"disclosureDate":"2025-12-08T13:17:27.000Z","capecIds":null,"crossRefCount":0,"attackSophistication":"moderate","impactType":["integrity"],"aiComponentTargeted":"model","llmSpecific":false,"classifierConfidence":0.85,"researchCategory":"peer_reviewed","atlasIds":null}}