ResembleAI 发布 DETECT-World 深度伪造检测模型
ResembleAI 推出了个新模型叫 DETECT-World，专门用来检测深度伪造，比传统方法更厉害。
ResembleAI 推出了 DETECT-World 模型，这是首个用于深度伪造检测的世界模型。传统检测方法会因新生成器而失效，而 DETECT-World 通过分析物理可能性（如光照、几何结构、运动和音视频同步）来识别伪造内容。
它在 Podonos 基准测试中音频准确率达到 99.5%，领先于其他商业 API。
We now have the world's best deepfake detector: @resembleai shipped DETECT-World, the first world model built for deepfake detection. Conventional detectors memorize known fakes. That degrades with every new generator, and Hugging Face now hosts 2.2M+ model variants. DETECT-World inverts the question: not "does this resemble a fake I've seen" but "is this physically possible." Lighting, geometry, motion, audio-visual sync. Physics doesn't ship a new version every week, so it catches generators it has never seen. One computation returns three things: a verdict, a timeline of when the anomaly occurs, and a heatmap of where. 99.5% audio accuracy on the externally run Podonos benchmark, first among commercial APIs. resemble.ai/learn/models/d…
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include ResembleAI, DETECT-World, deepfakes.
Source attribution: curated from AI Breakfast (@AiBreakfast).
