Google has quietly rolled out a new artificial intelligence system designed to significantly improve how fraudulent and policy-violating advertisers are detected within Google Ads.
The system, detailed in a Google research paper published in late December 2025, is already live in Google’s Ads Safety infrastructure and represents a major leap forward in both accuracy and reliability.
A New Model Built Specifically for Advertiser Risk
The new system is called ALF, short for Advertiser Large Foundation Model. Unlike previous approaches that relied on narrower signal sets, ALF is a multimodal foundation model built to evaluate advertiser behavior holistically.
It analyzes a wide range of inputs at once, including:
-
Ad creative content such as text, images, and video
-
Structured account data like account age and billing signals
-
Historical advertiser behavior and performance patterns
Google explains that while any single signal may appear harmless on its own, combining them reveals intent more clearly. Patterns that look normal in isolation can become highly suspicious when evaluated together.
Why Previous Fraud Systems Fell Short
Google outlined several limitations in earlier advertiser risk detection systems that ALF was designed to overcome.
First, advertiser data is highly heterogeneous and high-dimensional. Signals range from structured data points to unstructured creative assets, often numbering in the hundreds or thousands per advertiser. Traditional models struggled to process this breadth of information effectively.
Second, advertisers can upload massive volumes of creative assets. In some cases, malicious ads may be hidden among thousands of legitimate creatives, making detection difficult for systems that evaluate assets individually.
Finally, fraud detection systems must balance accuracy with trust. False positives can harm legitimate businesses, meaning confidence scoring must be reliable without requiring constant manual tuning.
Privacy-First Design
Despite using sensitive behavioral and account-level signals, Google emphasized that ALF was designed with strict privacy safeguards.
All personally identifiable information is removed before data is processed by the model. This allows ALF to identify risky behavior patterns without relying on personal or identifiable data.
How ALF Identifies Suspicious Advertisers
One of ALF’s key innovations is its use of a technique known as inter-sample attention.
Rather than evaluating advertisers in isolation, the model compares large groups of advertisers against one another. This allows ALF to learn what normal behavior looks like across the ecosystem and more accurately flag outliers that deviate from expected patterns.
By modeling relationships across advertisers, the system becomes more effective at identifying coordinated fraud, impersonation, and policy abuse.
Measurable Improvements in Real-World Performance
According to Google’s research, ALF significantly outperforms previous production systems.
In live environments, the model increased detection recall by more than 40 percentage points for certain policy violations while achieving precision rates as high as 99.8% on others. These gains were achieved not only in testing environments but in active production use, where ALF now handles millions of requests each day.
Google acknowledged that the model introduces slightly higher processing latency due to its size and complexity. However, the delay remains within acceptable thresholds for Ads systems and can be further optimized through hardware acceleration.
What This Means for Advertisers and Marketers
For legitimate advertisers, the rollout of ALF should result in a cleaner ad ecosystem with fewer fraudulent competitors and less policy abuse.
For marketers managing Google Ads accounts, the update reinforces the importance of maintaining compliant account structures, accurate billing information, and transparent creative practices. As Google’s detection systems become more sophisticated, patterns that once slipped through automated checks are increasingly likely to be identified.
Looking Ahead
Google noted that future iterations of ALF may incorporate time-based behavioral analysis to better detect evolving fraud patterns. The company also suggested potential applications beyond fraud detection, including audience modeling and creative optimization.
For now, ALF marks a significant step forward in how Google Ads evaluates trust, intent, and risk—using AI to protect both advertisers and users at scale.
