China's Ministry of Public Security (MPS) Decree No. 176, the Measures for Public Security Organs' Supervision and Inspection of Cyberspace Security, took effect on 1 October 2026. It was signed on 6 August and replaces the 2018 Decree No. 151. Neither the decree nor the facial recognition rules say anything about police using the technology on citizens. That omission is the policy problem.
What the decree does
According to Hunton Andrews Kurth's analysis, the measures cover network operators, critical information infrastructure (CII) operators, network product and service providers, and data and personal information handlers. Police can run online monitoring, including network patrols and vulnerability scanning. Prefecture-level authorities can run remote penetration tests on non-CII facilities. Capability testing and remote testing require three working days' notice. On-site inspections need at least two officers with credentials and a written notice, and routine visits to CII operators and high-tier multi-level protection networks are capped at one a year. Entities with earlier incidents or unremedied penalties get priority scrutiny.
The text on elawcn.com is a reprint of the decree. Article 13 lets police commission "cybersecurity service institutions or specialised personnel" to support inspections, under police direction, with notice filed to the next-higher police level. Those contractors sign confidentiality undertakings, and those doing penetration testing face background vetting. Article 7 makes compliance with data security and personal information protection duties an inspection item. I found no mention of biometrics or facial recognition in the reprint.
The case for the decree
The strongest argument for the decree is that 2018 rules written for "internet services" no longer fit a data economy. Ransomware, leaked databases and abused personal information are real harms, and Chinese citizens have seen their data traded openly. The decree adds some procedural discipline: credentials, written notices, annual caps for the most heavily regulated operators, record-signing with a place for objections, and liability for officers who leak or abuse information. Advisory letters for non-violation risks are also a softer tool than penalties. A regulator that can check whether companies actually protect personal data is better than one that cannot.
Where the design breaks down
The problem is structural. The March 2025 Measures for the Security Management of Facial Recognition Technology Applications were issued by the Cyberspace Administration of China (CAC) together with the MPS and took effect on 1 June 2025. Article 15 requires a filing with the provincial cyberspace authority once a handler stores facial information on more than 100,000 people. Article 13 restricts devices in public places to what public security requires and demands a clear defined area. Article 16 names cyberspace authorities and public security organs as supervisors.
The official text I reviewed contains no express exemption for state organs. Its only stated exclusion is research and algorithm training. But that is not the whole picture. The Personal Information Protection Law, as reproduced here, puts government bodies under a separate section. Article 34 says they may process personal information only within their statutory powers and procedures, and Article 35 lets them skip notification where it would impede their duties. The law therefore leaves wide discretion to the same ministry that is now the data-protection inspector. When the supervisor is also the largest user, supervision of the user is only as strong as the supervisor's willingness to apply it to itself.
The result is an asymmetry. A private company with a camera at a mall entrance faces filing, impact assessments, signage and, now, police inspection with contractors who can run penetration tests on its systems. A police bureau deploying the same cameras faces an internal chain of command. No independent body in this framework audits police use, and the decree gives affected people no route to challenge it. The EFF has described the same dynamic in the United States: agencies hiding surveillance technology from scrutiny because transparency invites pushback. China's version is less hidden than formally unaccountable.
A proportionate alternative
A pro-innovation, pro-rights approach would not scrap inspection. It would apply the same standards to everyone who runs the technology. Four changes would help:
- Require public-sector facial recognition deployments to be filed and published in summary, as private firms must already file above 100,000 records.
- Move audit of police-run systems to a body outside the police chain, such as the courts, procuratorate or an independent data protection authority.
- Narrow the contractor model with a public register of approved providers and an explicit ban on retaining inspected data.
- Give data subjects a way to ask whether and why their face data was collected.
These are not radical. They are the obligations China already imposes on companies, extended to the state. They also serve innovation. Firms in China's AI and camera industry are more likely to invest in privacy-protective design if the rules are credibly applied to everyone, including their largest customer.
What to watch
Two things will show whether the decree is a compliance tool or a surveillance tool. The first is whether inspections' stated focus on personal information produces published enforcement against any public-sector handler. The second is whether the 100,000-record filing regime ever reveals public-sector filings. Other jurisdictions are making similar choices, from Delhi Police's tender for facial recognition on all 338 CCTV cameras to the TSA safeguard findings in the United States. A law that protects biometrics from everyone except the party that most uses them will not build the public trust the technology needs.