Google on Wednesday released its Ad Safety Report for the year 2023 which says the company blocked or deleted over 5.5 billion advertisements and 12.7 million advertisement accounts that were in violation of its policies. The use of Generative AI, both as a tool for creating bad ads and for spotting them, was an overarching theme of the report. AI-based scam ads fought with AI Towards the eng of 2023 and into 2024, the company observed targeted campaigns of ads featuring the use of public figure likenesses to scam users. These ads often use deepfakes. The company detected patterns in these bad actors’ behaviours and trained their automated enforcement models to detect similar ads. Google also updated its misrepresentation policy to better cover the gambit of such ads that could enter its systems. Using generative AI to weed out scams Safety teams at the company have been using AI to enforce their policies at scale for years. That is how the company powers its efforts to detect and block billions of bad advertisements before they are seen by users. But Google says that these tools have needed extensive training on hundreds or thousands or even millions of examples of content that violates policy. But large language models can review and interpret content at a high volume while even “capturing nuances.” The advanced “reasoning” capabilities of the models helped Google take larger-scale and more precise enforcement decisions even with its most complex policies. The company served up the example of its policy against “unreliable financial claims,” which includes ads that promote get-rich-quick scams. The malicious actors behind these ads have grown more sophisticated as they adjust their tactics and tailor ads, positioning them as new financial services or products like investment advice or digital currencies. Yes, conventional machine learning models are trained to detect these policy violations. But their work is cut out for them because of the fast-paced and ever-evolving nature of financial trends. This makes it more difficult to differentiate between legitimate and fake service capabilities. But Google says LLMs are more capable of quickly recognising trends in financial services and identifying patterns of bad actors who are abusing these trends. What Google is doing to prevent fraud and scams Overall, Google blocked or removed 206.5 million advertisements for violating its misrepresentation policy (which includes scam tactics). 273.4 million advertisements were banned for violating its financial services policy while it blocked or removed over 1 billion advertisements for abusing its ad network, including those that promote malware. Election advertisements According to Google, all election ads should include a “paid for by” disclosure and it verified more than 5,000 new election advertisers and removed more than 7.3 million election ads that came from advertisers who did not complete verification.
