AppleStorm: Unmasking the Privacy Risks of Apple Intelligence
Analyzes the architecture of Apple Intelligence and Private Cloud Compute, identifying privacy risks and data leakage vectors across on-device and cloud models.
Summary
Apple Intelligence and Private Cloud Compute
Apple Intelligence integrates on-device generative AI models with Private Cloud Compute (PCC) to execute complex natural language tasks while maintaining strict user privacy guarantees.
Architectural Analysis of On-Device AI Routing
Yoav Magid evaluates the client-side decision-making algorithms that route user prompts and context between local neural engines and remote cloud compute clusters.
Data Exposure and Cross-App Context Leakage
Security analysis uncovers edge-case vulnerabilities where semantic indexing pipelines and app intent handlers inadvertently expose sensitive user context across unprivileged third-party apps.
Hardening Confidential AI Architectures
Ensuring absolute privacy in hybrid AI architectures requires verifiable remote cryptographic attestation, strict on-device data redaction, and independent security auditing of serverless inference nodes.
Related talks
Main Stage·45 min
Richard Im
Uncovers privacy vulnerabilities and local IPC leakage across Apple's Siri, CloudKit, and system daemon ecosystems on iOS and macOS.
Black Hat USA·36 min
Austin Emmitt
Engineers explore architecture and security mechanisms for Apple's Predicament: NSPredicate Exploits on iOS and macOS.
Creator Stage·26 min
Andrea Downing
Introduces the Patient AI Rights framework and a community-driven red teaming methodology to stress test clinical AI models and chatbots for bias and failure modes.