MobileGuard: A Mobile-Native Governance Framework for Agentic AI
MobileGuard is the first mobile-native governance framework for agentic AI, designed to address structural constraints of consumer mobile platforms. Validated through three studies, it reduced deployment errors by 74.1% and identified 23 failure categories, with 71.3% undetectable by existing frameworks. An audit of 942 apps found a 4.0% governance signal rate, including violations by major developers. The framework maps to ISO 42001:2023 and the EU AI Act, and is available as open-source.
Published June 27, 2026
| Version v1
Preprint
Open
MobileGuard: A Mobile-Native Governance Framework for Agentic AI
Authors/Creators
Singh, Jaspreet (Researcher)
Description
Consumer mobile platforms now constitute the primary delivery channel for agentic AI, with global app releases surging 60-104% year-over-year in 2026, driven by AI-assisted development tools. Yet existing agentic governance frameworks, designed for mutable server-side enterprise deployments, cannot address the structural constraints unique to mobile: binary immutability, platform gatekeeper non-determinism, consumer-scale blast radius, ambient agent surface expansion, and regulatory exposure. We present MobileGuard, the first mobile-native governance framework for agentic AI, operationalizing four pillars across the mobile SDLC: Pre-Deployment Quality Contracts (PDQC), Tiered Autonomy Calibration for Mobile (TAC-M), Platform Gatekeeper Simulation and Governance (PGSG), and Ambient Agent Boundary Enforcement (AABE). MobileGuard is validated through three empirical studies. Study 1 presents a governance failure taxonomy derived from 2,847 real-world iOS and Android platform rejection records, identifying 23 failure categories across 6 pillars with 71.3% of failures undetectable by existing frameworks. Study 2 demonstrates a 74.1% Deployment Error Rate reduction (p less than 0.001) against a production mobile SDLC pipeline across three applications, outperforming AGENTSAFE by 45.7 percentage points. Study 3 presents a real-world cross-sectional audit of 942 mobile platform applications (iOS-primary; Android replication in progress) using AS-009, a novel release notes AI disclosure scanner covering both App Store and Google Play governance signals, finding a 4.0% governance signal rate with externally-derived violations observed in enterprise-scale developers including Adobe Inc. and Moleskine Srl, resolving the author-defined scenario circularity concern from Study 2. MobileGuard maps to ISO 42001:2023 and the EU AI Act, and is implemented as an open-source CLI available at github.com/jsingh6/mobileguard.
Files
MobileGuard.pdf
Files (306.6 kB)
Name Size
Download all
MobileGuard.pdf
md5:acb723f8265c0216f931caf9e66b2a14
306.6 kB
Preview
Download
Additional details
Software
Repository URL
https://github.com/jsingh6/mobileguard
Programming language
Python
Development Status
Active