You Still See Me: How Data Protection Supports the Architecture of AI Surveillance

Privacy-preserving techniques can paradoxically enable more surveillance: The paper demonstrates that methods like homomorphic encryption and federated learning—often marketed as privacy protections—can actually facilitate expanded data extraction and surveillance when deployed within AI systems, despite appearing compliant with data protection regulations.

Compliance rhetoric masks surveillance infrastructure: Organizations may use privacy and data protection law compliance as cover to justify surveillance AI systems, creating a false sense of regulatory safety while the underlying surveillance architecture remains intact and potentially more invasive.

Current privacy protections are insufficient for AI contexts: The paper argues that existing data protection frameworks don't adequately account for how privacy-preserving techniques are applied across the full AI development pipeline—from dataset creation through model training—leaving significant gaps in actual protection.

Technologists have responsibility in policy design: The authors emphasize that technology developers should play an active role in creating policies that genuinely combat surveillance AI, rather than leaving policy decisions solely to regulators, suggesting that technical expertise is essential for effective governance.

We need new evaluation frameworks: The paper calls for technology and policy strategies to assess whether privacy-preserving techniques actually deliver the protections they claim, implying current evaluation methods are inadequate for determining real-world privacy outcomes.