Subho Halder

AI Summary

→ WHAT IT COVERS Subho Halder explains how mobile app security fails to keep pace with AI-driven development cycles, where fake applications harvest user data from app stores, traditional penetration testing becomes obsolete, and trust erosion forces companies to prove transparency in data handling as AI agents create new attack vectors requiring automated defense systems. → KEY INSIGHTS - **Fake Application Categories:** Three types of malicious apps infiltrate app stores: ad-revenue wrappers that monetize popular AI brands, data farming apps that harvest contacts and location for sale to brokers and competitors, and malware that directly attacks users. Play Store and App Store remove 200,000-250,000 apps annually, with 50-60% violating safety norms, yet benign-appearing data collectors evade detection. - **AI Democratizes Hacking:** Script kiddies evolved into prompt engineers who instruct AI models like Claude to compile Android apps and identify vulnerabilities without understanding code. This lowers the barrier to entry for attackers while defenders must cover all bases since attackers need only one failure point. The reasoning capability of AI models transforms offensive security from algorithmic pattern matching to adaptive threat generation. - **Developer Burnout Shifts:** AI generates code in minutes but creates review bottlenecks lasting days. Developers spend hours understanding AI-generated pull requests without human authors to consult, reviewers use additional AI tools causing confusion, and QA teams test code with unknown intent. The fatigue moved from writing code to validating and deploying it, not eliminating the problem but relocating it downstream in development cycles. - **Trust Requires Transparency:** Companies build trust through three mechanisms: transparent data processing explanations, certifications like SOC 2 Type 2 that enforce access controls, and government accountability where regulators can summon companies. Users trust OpenAI over DeepSeek because US Congress can hold domestic companies accountable for data breaches, while foreign entities operate beyond jurisdictional reach, making trust psychological rather than purely technical. - **Mobile Holds Concentrated Risk:** Mobile devices store credit cards, SSNs, healthcare records, and behavioral data while security treats them as thin clients assuming server-side risk. Apps request one-time permission grants during installation that users ignore, unlike desktop browsers that prompt per-session. Release cycles compressed from months to days, APIs multiplied exponentially, and third-party SDKs exploded, transforming apps from static products into living systems requiring behavioral security models. → NOTABLE MOMENT Halder reveals AppKnox deployed an AI agent that automatically detects API errors, identifies code locations in GitHub, and submits pull requests for fixes. The system generates code in one minute, but developers require a full day to understand the root cause, reviewers struggle without human authors to consult, and senior engineers must intervene before production deployment, demonstrating how automation relocates rather than eliminates workload. 💼 SPONSORS [{"name": "Tastytrade", "url": "https://tastytrade.com"}] 🏷️ Mobile App Security, AI Penetration Testing, Fake Applications, Developer Burnout, Data Privacy