China AI Trade Secret Law: Algorithms and Datasets Protected June 1
Abhishek Gautam10 min readQuick summary
China SAMR trade secret rules effective June 1, 2026 classify AI algorithms, datasets, and code as protected secrets. Fines up to 5M yuan; strict cross-border access logs.
Read next
- Anthropic Offers EU ENISA Access to Mythos After San Francisco Talks
- OpenAI Launches Rosalind Biodefense for GPT-Rosalind Life Sciences AI
China's "Trade Secret Protection" regulations took effect June 1, 2026 — the first update since 1998 — explicitly classifying AI algorithms, datasets, and computer code as trade secrets when not publicly disclosed, with SAMR fines up to 5 million yuan (~€630,000) and mandatory access logs for cross-border collaboration.
The rules land the same week as outbound investment controls (July 1 rules) and Meta Manus fallout — Beijing is locking AI IP, people, and data inside its regulatory perimeter.
What Counts as a Trade Secret Now?
Trade secrets under the June 2026 rules include technical information that is:
- Not publicly known
- Has commercial value
- Is kept confidential
Explicitly listed: structure, raw materials, formulas, processes, methods, data, algorithms, computer programmes, and codes.
Already public information (media, open reports) is excluded.
| Requirement | Detail |
|---|---|
| Remote / cross-border work | Access controls, PII anonymization, audit logs (who accessed, copied, modified) |
| Enforcement | SAMR investigations; fines up to 5M yuan |
| Awareness campaign | June 2026 "Enterprise Trade Secret Protection Capacity Enhancement" month (annual since 2023) |
Why This Matters for Developers Outside China
1. API and distillation risk
Systematic extraction of model behavior may now be pursued as trade secret violation domestically — parallel to US Economic Espionage Act framing in White House AI theft memo.
2. Cross-border R&D contracts
Teams with China-based contractors need logging, data residency, and export clauses aligned with SAMR and July outbound rules.
3. Open-weight ambiguity
Weights published openly may fall outside trade-secret scope; training data, fine-tunes, and undisclosed eval harnesses may not.
4. Talent movement
Bloomberg reported restrictions on AI talent movement the same month — trade secrets + outbound rules + Manus travel limits form one policy arc.
Pair with China blocks Meta Manus and DeepSeek on domestic chips.
For compliance tooling, see LLM API Pricing and Claude vs ChatGPT when choosing where inference runs.
Key Takeaways
- June 1, 2026: China Trade Secret Protection rules now cover AI algorithms, datasets, code
- SAMR can fine up to 5 million yuan for confirmed leaks
- Cross-border work requires access controls, anonymization, detailed logs
- First update since 1998; part of same-week outbound investment tightening
- For developers: treat undisclosed Chinese model assets like export-controlled IP in contracts and API use
Sources
FAQ
Frequently Asked Questions
When did China AI trade secret rules take effect?
China's Trade Secret Protection regulations took effect on June 1, 2026. They are the first update to China's trade secret framework since 1998 and explicitly include algorithms, datasets, and computer code.
What is the maximum fine for trade secret violations in China in 2026?
China's State Administration for Market Regulation can impose fines of up to 5 million yuan, approximately €630,000, if an investigation confirms trade secrets were divulged.
Are AI algorithms considered trade secrets in China now?
Yes. The June 2026 rules state that information related to technology including data, algorithms, computer programmes, and codes can qualify as trade secrets if not publicly known, commercially valuable, and kept confidential.
How do China trade secret rules affect cross-border AI teams?
The regulations require strict measures for remote and cross-border collaboration, including access controls, data anonymization for personal information, and detailed logs of who accessed, copied, or modified protected information.
Free Weekly Briefing
The AI & Dev Briefing
One honest email a week — what actually matters in AI and software engineering. No noise, no sponsored content. Read by developers across 30+ countries.
No spam. Unsubscribe anytime.
More on AI Models
All posts →Anthropic Offers EU ENISA Access to Mythos After San Francisco Talks
Anthropic invited EU cyber agency ENISA to access Mythos via Project Glasswing after Commission talks in San Francisco. Terms still pending; EU AI Act August 2026.
OpenAI Launches Rosalind Biodefense for GPT-Rosalind Life Sciences AI
OpenAI announced Rosalind Biodefense on May 29, 2026 — sponsored GPT-Rosalind access for vetted developers building pandemic detection, screening, and countermeasure tools.
Iran Black Shadow Used ChatGPT to Wipe US, Mideast IT Systems
Gambit Security links Iran MOIS group Black Shadow to June 2026 destructive campaign in US, Israel, Saudi Arabia, Turkey. Attackers used ChatGPT to refine wipe scripts.
Gemini 3.1 vs Claude Sonnet 4.6 vs GPT-5.3 Codex: Developer Benchmark Comparison March 2026
Gemini 3.1 Pro, Claude Sonnet 4.6, and GPT-5.3 Codex all dropped within weeks of each other in early 2026. Here's how they actually compare on coding benchmarks, context windows, API pricing, and which model to use for what — a developer-first breakdown with real numbers.
Free Tool
Will AI replace your job?
4 questions. Get a personalised developer risk score based on your stack, role, and what you actually build day to day.
Check Your AI Risk Score →Written by
Software Engineer based in Delhi, India. Writes about AI models, semiconductor supply chains, and tech geopolitics — covering the intersection of infrastructure and global events. 795+ posts cited by ChatGPT, Perplexity, and Gemini. Read in 164 countries.