Privacy Computing in Digital Government: Background, Technical Roadmap, Case Studies, Challenges, and Recommendations

Background Introduction Since 2018, concepts such as data usability without visibility have emerged, and the digital government sector initially promoted trusted data sharing using sandbox technologies. Subsequent policies, including the Public Data Resource Development and Utilization Pilot Program and the 14th Five‑Year Plan, have accelerated the privacy‑computing industry, making sandbox techniques relatively less advanced than multi‑party computation and federated learning.

Legal regulations and data application rules across regions aim to resolve data ownership and security issues. Privacy computing ensures data ownership during computation, maximising data value while providing safety for both data providers and consumers.

Technical Roadmap

1. Privacy Computing Technologies – Governments may adopt different techniques based on data sensitivity: low‑sensitivity data can use trusted execution environments (TEE) or secure sandboxes; medium‑sensitivity data may employ multi‑party secure computation; high‑sensitivity data might require federated learning combined with homomorphic encryption.

2. Multi‑Party Secure Computation – Relies on underlying cryptographic protocols for trust.

3. Federated Learning – Guarantees that local data never leaves its source.

4. Trusted Execution Environment – Trust is placed in the chip and the sandbox it creates.

5. Secure Sandbox – Provides a closed, safe, and free computing environment. Deployment decisions (government side, bank side, or distributed) affect where the sandbox resides and how results are handled.

Case Sharing

1. Internal Government Data Sharing – Example: a natural resources department needed to verify officials' real‑estate holdings without revealing the queried individual. A privacy‑computing solution introduced query obfuscation (mixing multiple records) to protect both the query and the data.

2. Government‑Enterprise Data Utilisation – Banks require government data for credit scoring. After personal data protection laws, data is accessed via data‑bureaus that protect raw data using privacy‑computing platforms, delivering processed credit scores to banks while preserving data ownership.

3. Public Data Operation Platforms – Various provinces authorize state‑owned enterprises to operate public data platforms, offering data services, but the number of privacy‑computing products remains limited due to unclear product definitions.

4. Data Exchanges / Centers – Beijing pioneered a 3.0 data‑exchange model incorporating privacy computing; Shanghai and other regions have followed with varying degrees of adoption.

Implementation Challenges

Economic constraints: pilot projects have limited budgets, and high development and verification costs often lead to losses.

Security assurance: proving platform safety for multi‑party computation or federated learning lacks authoritative certification.

Practical scenario gaps: vendors focus on platform development without sufficient business expertise, making coordination among government units, banks, and research institutes difficult.

Platform performance and usability: low technical maturity in government units results in complex, low‑adoption platforms.

Suggested Approaches

Enhance data‑management capabilities to integrate with existing government information systems.

Implement data classification and grading to select appropriate privacy‑computing techniques.

Develop unified data‑service development tools that bridge various underlying computation platforms.

Improve data supply‑demand matching beyond simple request forms.

Strengthen security governance and compliance review, recognising that manual audit remains essential.

Deploy data‑security gateways and consider blockchain contracts for auditability and fault tolerance.

Q&A Highlights

Q1: Future development of domestic privacy‑computing companies?

A1: Most vendors are shifting toward data‑service offerings to provide concrete solutions for government clients, while a minority focus on open‑source technology development.

Q2: How does MD5 comparison help assess data quality in the second case?

A2: Banks compare their own stored data with government data using MD5 hashes to verify consistency; however, this does not fix underlying data‑quality issues.

Q3: Does data operation risk personal privacy leakage, and how is it protected?

A3: Personal data must be authorized; without explicit consent, governments generally do not expose personal information to third parties.