Using AI to write code is now routine in R&D: humans design the architecture, AI generates the code, humans revise. Which raises an increasingly urgent question — can such software still be registered for copyright, and still support a High-Tech Enterprise application?

Three takeaways for companies and fellow practitioners:

1. Registration got stricter in 2026.

China's Copyright Protection Centre (CPCC) has shifted software copyright registration to a limited substantive examination — cross-checking function, code, and documentation — and introduced an integrity regime: applicants must declare that the software was “independently developed by humans, without AI writing the code.” A false declaration can land you on a credit blacklist, synced to the national credit system. Where AI is deeply involved, signing the old declaration may amount to a false filing — with enterprise-level consequences.

2. The law actually accepts AI assistance.

Since November 2023, Chinese courts have consistently treated AI as a tool: so long as a human makes an “original intellectual contribution” to the final expression, the work remains protectable. The tension is this — courts ask how much the human contributed (a question of substance); the registration form asks whether AI was used at all (a binary). The answer isn't clever wording. It's making human-led development a fact, and documenting it.

3. Structure your protection — layered, and genuine.

This matters most for High-Tech Enterprise certification (a Chinese tax-incentive status), where software copyrights are Class II IP — usable only once and scored lower, so one invention patent is worth roughly five or more software copyrights. Don't pad the count with batch, AI-generated registrations. Instead:

– Core methods → invention patents (they protect the technical solution and best withstand scrutiny)

– Non-public know-how → trade secrets

– Specific code expression → software copyright (register only what genuinely reflects human authorship)

– And throughout: a full Git trail, open-source and ownership checks, function/code/documentation consistency, and registration timing aligned with your R&D projects

The value of IP in the AI era lies in putting human intelligence into architecture and judgment, keeping AI in the tool's seat — and proving it with evidence.