New LLM-Based Tool Detects Hidden Ad Regulation Violations in Mobile Apps

Mobile app developers must navigate complex advertising regulations, but enforcement remains difficult because ads are fetched from remote sources and placed across diverse screens at runtime. Researchers have now introduced a new approach to find these violations at scale.

The Challenge#

Detecting problematic mobile ads requires examining both an app's source code and its actual runtime behavior. Ads are often loaded dynamically from external servers and embedded throughout various user interface elements, making them hard to locate through code review alone. Existing detection methods struggle with this complexity, allowing many regulation violations to slip through undetected.

How AdWise Works#

The researchers built AdWise, a system that harnesses large language models to combine static code analysis with dynamic app exploration. The approach operates in three stages:

• Static analysis: Identifies UI components designed to display advertisements (ad widgets) by examining the app's code
• Dynamic exploration: Uses an LLM to navigate through the app's screens and trigger ad displays, guided by three complementary strategies
  • Window transition graphs map the app's structural layout
  • Semantic reasoning prioritizes paths users actually visit
  • Retrieval-augmented generation (RAG) recognizes common ad-heavy interface patterns from a knowledge base

Results and Real-World Impact#

Testing on 100 benchmark apps showed AdWise outperformed competing approaches by 25.83 percentage points in detecting ad-placement widgets. The system uncovered 34.34% more regulation violations across six violation categories spanning three regulatory frameworks.

In a broader study of 2,047 apps using Meta advertisements, researchers identified 383 compliance violations. Findings were reported to both Google Play and Meta's bug bounty program.