Google Play 使用 AI 於 2025 年阻擋 175 萬惡意應用程式

• Google's AI system employs multi-layer protections combining machine learning models trained on app behavior analysis, code pattern recognition, developer history evaluation, and real-time user interaction data[2] • The system analyzes financial fraud schemes, spyware disguised as legitimate utilities, and malicious code patterns to make real-time decisions on app approval[2] • Play Integrity API processes over 20 billion checks daily, with 2025 enhancements including new hardware-backed signals and in-app remediation prompts to protect apps against abuse and unauthorized access[3] • Enhanced fraud protection blocks risky installation attempts from internet sources (browsers, messaging apps) that request sensitive permissions, now deployed across 185 markets[4] • Generative AI models are integrated into the human review workflow to accelerate detection of complex and evolving malicious patterns, reducing time-to-detection for sophisticated threats[3][5]

Google's 2025 security achievements signal a maturation of AI-driven app store governance, with the 25% year-over-year decline in blocked apps suggesting that proactive deterrence is reshaping developer behavior rather than simply catching more threats. This trend may establish a new industry standard where AI-powered prevention becomes the primary defense mechanism, potentially influencing how competitors like Apple and Microsoft approach app store security. The scale of Google's operations—350 billion daily scans across 3 billion Android users—demonstrates AI's capacity to deliver measurable consumer protection at unprecedented scale, positioning machine learning as essential infrastructure for digital ecosystems. However, the decline also raises questions about whether malicious actors are adapting by shifting to alternative distribution channels (sideloading, third-party stores) rather than being eliminated entirely, suggesting the security arms race will continue to evolve. Google's emphasis on developer accountability through verification requirements and mandatory pre-review checks indicates a shift toward ecosystem-wide responsibility rather than reactive threat detection alone.