Filed under de-identification
Every brief tagged "de-identification".
- § 01 · ANONYMIZATION
Reviving a Zombie Provision — Xu Ke's Concentric-Circle Reconstruction of the Anonymization Regime
Xu Ke (UIBE) calls PIPL Article 4's anonymization carve-out a 'zombie provision' (僵尸法条) — on the books, never used, and one of the biggest blockages in the data-element market. His diagnosis: the zombie state is caused not by the text but by three unaddressed worries (processors fear the standard is unattainable or value-destroying; regulators fear anonymization becomes an evasion tool; users fear it's a hollow promise). His cure is a concentric-circle architecture that maps three risk types (systemic / operational / residual) onto three layers of anonymity (presumptive / determined / trust). This is the most complete academic blueprint yet for making the anonymization clause operational — and it pairs directly with TRIMPS's risk-based, recipient-relative reading.
- § 02 · ANONYMIZATION
From 'Cannot Be Restored' to 'Difficult to Restore' — TRIMPS on Whether Anonymization Is Absolute, and Whether It's Recipient-Relative
The Third Research Institute of the Ministry of Public Security (TRIMPS) — the body behind China's classified-protection regime and national eID platform — takes on the two questions that determine whether anonymization actually gets data out of PIPL scope. First: does PIPL's 'cannot be restored' standard (Art 73) require re-identification probability of literally zero? The 2025 draft PI Anonymization Guide quietly softened it to 'difficult to restore,' aligning China with the GDPR 'all reasonable means' test and reframing anonymization as a dynamic, continuously-assessed, risk-based process rather than a one-time terminal state. Second: is anonymization recipient-relative — can the same dataset be PI in one party's hands and anonymized in another's? TRIMPS reads the EU SRB v EDPS case and UK ICO guidance toward 'yes,' with major implications for how overseas counsel structure data sharing and cross-border transfer.