在生成式人工智能技术迅速发展的背景下,达特茅斯学院教授Hany Farid主张将数字媒体真实性验证的默认立场从传统的“假定真实”转变为“假定虚假”,以应对大规模内容造假带来的挑战 1。这一观点源于Farid对伊朗Mehr News机构发布的一段声称展示美国Tomahawk导弹袭击学校视频的分析;尽管其结论是该视频为真实影像,但此类案例凸显了传统法证手段在面对深度伪造时的局限性 1。
Farid指出,鉴于加州大学伯克利分校的一项研究显示观察者因普遍怀疑心理而难以确信视频真实性,社会亟需建立新的信任机制 1。他建议通过推广C2PA标准、嵌入Google SynthID等隐形水印以及利用硬件签名技术来赋予创作者证明内容真实的工具 1。例如,苹果最新款iPhone已实现拍摄时对原始像素数据进行加密签名,从而生成可用于后续比对的参考图像 1。
除了静态媒体,实时通信中的深伪风险同样受到关注。Farid特别提及2024年9月参议员Ben Cardin遭遇疑似由俄罗斯策划的、冒充乌克兰前外长Dmytro Kuleba的深度伪造视频通话事件 1。为此,他呼吁未来的视频通话软件应引入与静态照片类似的帧级加密验证机制,以确保摄像头传感器的物理来源可信 1。
In an era where generative artificial intelligence makes fabrication increasingly accessible, experts are advocating for a fundamental shift in how society evaluates visual evidence. Hany Farid, a professor at Dartmouth College, argues that the default stance toward recordings should move from assuming they are authentic to presuming they are fake until proven otherwise 1. This proposed paradigm change aims to transfer the burden of proof from skeptical viewers to content creators, requiring them to actively demonstrate the authenticity of their media 1. Farid’s perspective is grounded in forensic analysis and emerging technical standards designed to counter mass-scale manipulation that traditional verification methods can no longer adequately address 1.
The urgency of this approach is illustrated by recent geopolitical events and security breaches. Farid analyzed a video released by Iranian news agency Mehr News, which claimed to show U.S. Tomahawk missiles striking a school; his forensic examination concluded that the footage was indeed genuine 1. However, such verifications are becoming harder to sustain amidst widespread skepticism, as research from the University of California, Berkeley indicates that observers struggle to trust videos due to pervasive doubts about digital integrity 1. The risks extend beyond recorded files to real-time communications, exemplified by a September 2024 incident involving Senator Ben Cardin, who received a video call featuring a deepfake of former Ukrainian Foreign Minister Dmytro Kuleba, suspected to be orchestrated by Russian actors 1.
To combat these threats, the industry is turning to cryptographic solutions and hardware-based authentication. Proposed mechanisms include adopting the C2PA standard, which attaches encrypted signatures to verify content provenance, or embedding invisible watermarks like Google’s SynthID 1. Hardware innovations also play a critical role, with Apple’s latest iPhone capable of encrypting raw pixel data at the moment of capture to create a reference image for future comparison 1. Looking ahead, experts suggest that video calling software must implement frame-level encryption similar to that used for static photos, ensuring that live feeds can be traced back to their physical camera sensors to prevent deepfake fraud 1.
评论
还没有评论,欢迎留下第一条。