Cybersecurity Multi-Level Protection
MLPS is not a one-off assessment but an obligation chain running from grading through the system's whole life. Most AI companies first hear about it when a tender blocks them.
MLPS and algorithm filing are two different things
Algorithm filing governs whether an algorithm service may be offered; MLPS governs whether the system carrying it is secure. Different object, different authority, different materials — yet the two are routinely confused, until an enterprise customer demands MLPS filing evidence during procurement and the work has never been done.
MLPS 2.0 has five steps: grading, filing, remediation, assessment and supervision. Operators grade their own systems with expert review; level 2 and above must be filed with public security. Level 3 systems are generally assessed annually, and failures require remediation and re-assessment.
A common misjudgement among AI companies is assuming “we are only SaaS, level 2 is enough”. Grading turns on the harm to national security, public order, public interest and the rights of citizens and organisations if the system is compromised — user scale, data sensitivity and business continuity all matter. An inference platform carrying large volumes of user data often grades higher than the team expects.
What we do
- System and data inventory, grading: determine the level and its basis
- Filing: filing forms, grading report, expert review opinions
- Remediation: gap analysis against level requirements, technical and management plans
- Assessment support: assessor liaison, issue remediation, re-assessment
- CII identification: assessment of critical information infrastructure exposure and resulting obligations
- Institutional readiness: security policies, incident plans, drills and training records
Typical situations
Situation 1: Tender requires MLPS filing evidence, three weeks before deadline
A government tender requires level 3 filing evidence. The full cycle of grading, filing, remediation and assessment normally takes months.
Our approach: grade in advance based on the business plan and file early for systems intended for public-sector work; for short-term bids, adjust system boundaries and scope to match an achievable level rather than forcing level 3.
Situation 2: Level 3 assessment failed and renewal stalls
The system is graded level 3. Assessment finds deficiencies in identity authentication, access control and log audit. The customer’s annual supplier review requires a pass certificate and renewal is suspended.
Our approach: remediate item by item, prioritising high-risk findings, produce a remediation report and schedule re-assessment; build routine self-checks rather than a pre-assessment scramble.
Situation 3: Designated as critical information infrastructure
After user growth the platform is brought within CII scope, with stricter obligations including a dedicated security body, annual testing and incident reporting. The company had built to level 3 and still falls short.
Our approach: re-assess against CII requirements, establish the dedicated security function and responsible person, and reset testing frequency and reporting.
Process and deliverables
| Stage | Work | Deliverables |
|---|---|---|
| 1 | System and data inventory, grading | Grading report + expert review |
| 2 | Public security filing | Filing certificate |
| 3 | Gap analysis and remediation | Remediation plan + records |
| 4 | Assessment | Assessment report + re-assessment plan |
| Ongoing | Annual self-check, drills, policy updates | Self-check report + drill records |
Frequently asked
Should MLPS and algorithm filing be done together? They can run in parallel but are not the same. Algorithm filing goes to the cyberspace administration; MLPS goes to public security. Materials, timelines and owners differ, so starting both together compresses the overall schedule.
Level 2 or level 3 — who decides? The operator makes the initial determination, level 2 and above requires expert review, and the filing authority’s view prevails. Under-grading is corrected at assessment or inspection.
We run on the cloud — is MLPS the provider’s responsibility? No. The cloud provider carries obligations for its platform, but enterprises must still grade and file their own business systems. Responsibility does not transfer with hosting.
What happens if assessment fails? Rectification orders, warnings, fines — and, more immediately, blocked procurement and renewal. Commercial impact usually arrives before penalties.
Legal basis
- Cybersecurity Law
- MLPS regulations and the MLPS 2.0 national standard series
- Regulations on the Security Protection of Critical Information Infrastructure
- Data Security Law and PIPL security obligations
- Measures for the Administration of MLPS Assessment Institutions