谷歌发布2024年广告安全报告，重点介绍了其利用先进的大型语言模型（LLMs）检测和打击广告欺诈的行为。通过对LLMs的50项增强，谷歌能够更快速地识别新兴威胁、识别滥用模式，并将合法商家与诈骗区分开来。这导致了去年51亿条广告被拦截或移除，并暂停了3920万个欺诈账户。谷歌还利用AI打击利用AI进行诈骗的恶意行为者，特别是深度伪造。通过封禁超过70万个利用AI生成图像或音频的广告商账户，相关报告下降了90%。

🚫 谷歌利用AI技术，增强了其大型语言模型（LLMs），以更有效地检测和打击广告欺诈。这些增强包括50项改进，使其能够更快地识别新兴威胁和区分合法商家与诈骗。

📊 谷歌在2023年拦截或移除了51亿条广告，主要原因是滥用广告网络，包括通过诱饵式广告或恶意软件规避谷歌的审查流程。此外，谷歌还处理了商标侵权和违反个性化广告政策的行为。

🎭 谷歌利用AI对抗利用AI进行诈骗的恶意行为者，特别是深度伪造。例如，针对利用AI生成图像或音频冒充名人进行诈骗的行为，谷歌暂停了超过70万个广告商账户，相关报告下降了90%。

🛡️ 谷歌广告安全团队在用户看到广告之前就关闭了大多数欺诈账户，这显示了其在保护用户方面的主动性。

Google's use of AI to combat harmful ads has resulted in the suspension 39.2 million fraudulent accounts.

On Wednesday, the tech giant published its 2024 Ads Safety Report, highlighting its use of advanced LLMs (Large Language Models) to detect and enforce advertiser fraud. Since 2023, Google has added "50 enhancements" to its LLMs that "need only a fraction of the information earlier models needed to quickly recognize emerging threats, identify patterns of abuse, and distinguish legitimate businesses from scams." Indicators of abuse that can be detected by Google's AI tool include business impersonation and illegal payment details.

As a result, Google blocked or removed 5.1 billion ads last year. The majority of those ads Google stopped were due to "abusing the ad network," meaning methods of circumventing Google's review process by tricking users with bait-and-switch ads or using malware. Other ads caught in the sweep involved trademark violations as well as personalized ads that violate Google's policies by targeting users or promoting products based on sensitive topics like personal hardships, identify and belief, and sexual interests.

Google is also using AI to fight the increase of bad actors leveraging AI for scams. Deepfakes have become more pervasive and convincing. Last year, actor Tom Hanks' likeness was used to shill medical hoaxes. Scarlett Johansson took legal action against an app for deepfaking her image and voice to promote it. Google went after "bad actors using AI-generated imagery or audio to imply an affiliation with a celebrity to promote a scam" by suspending over 700,000 advertiser accounts, which led to a 90 percent decrease in reports. Overall, it blocked or removed 415 million scam ads.

Google's ad safety team said they shut down the majority of scammy accounts before users were ever served an ad. Given the present volume of offensive of harmful stuff on the internet, we shudder to think of the ads that never saw the light of day.