Basic Security Requirements for Generative Artificial Intelligence Services (TC260 technical document)

生成式人工智能服务安全基本要求

Record CN-06 · TC260-003

Issued by: National Cybersecurity Standardization Technical Committee (TC260)
Jurisdiction: China
Level · region: National · East Asia
Published / adopted: 2024-02-29
Language: Chinese
Official identifier: TC260-003

Summary

TC260 technical document setting the basic security requirements that generative AI service providers must follow, in support of the Interim Measures for generative AI services. It covers corpus security (sources with over 5% illegal and unhealthy information are excluded), model security, security measures, keyword libraries and test question banks, and the security assessment that providers report on when they complete filing procedures. Appendix A lists 31 main security risks of corpora and generated content. Providers must also watch long-term risks such as AI able to deceive humans, self-replicate or self-modify, and its use for malware or biological or chemical weapons.

Language

English translation by Center for Security and Emerging Technology (CSET), Georgetown University, not by a public body Open it. The Chinese text is authoritative.

Safety topics

general-purpose / frontier AIevaluation & testingcatastrophic risk

Status notes

Released 29 Feb 2024; a preliminary draft had been published in October 2023. It was designed to inform a national standard, and its content was taken up in the recommended national standard GB/T 45654-2025 (CN-10), published 25 Apr 2025 and effective 1 Nov 2025.

Sources

Open the official source https://www.tc260.org.cn/portal/article/2/20240301164054

Secondary source: https://cset.georgetown.edu/publication/china-safety-requirements-for-generative-ai-final/

Local copy

Archived, needs review. The page barely mentions the document's title; check that the URL still points to this document. A source changed since the previous snapshot; both versions are kept. Last attempt 2026-10-14 10:16:24 UTC

Official source (third-party copy, not published) archive/CN-06/2026-10-14/official.html

Retrieved 2026-10-14 07:25:16 UTC

SHA-256 b882626d90f68a77283c19776a2fd8ca948768c4721810a260c81965c0a9aa59

Warning: The page barely mentions the document's title; check that the URL still points to this document

Official source, rendered in a browser (third-party copy, not published) archive/CN-06/2026-10-14/official-rendered.html

Retrieved 2026-10-14 07:26:20 UTC

SHA-256 1dcdce3448e9fe6df8d83b050ea1c26af331d9d35cd24f11880623798229abe5

Warning: The page barely mentions the document's title; check that the URL still points to this document

Official source, printed to PDF by a browser (third-party copy, not published) archive/CN-06/2026-10-14/official-print-4.pdf

Retrieved 2026-10-14 10:23:47 UTC

SHA-256 035b263fb8c4de40670bb975bbd73533acbfb03f175c889e28f7eecf11ac0e46

Secondary source (third-party copy, not published) archive/CN-06/2026-10-14/secondary.html

Retrieved 2026-10-14 07:25:20 UTC

SHA-256 3eec0c742e05510ff62977a9bcf02344f5fc06a9ae62b1e6648495414688e455

Verification

Not confirmed on an official page; the record rests on secondary sources. Check it before citing. checked 2026-10-14 · search-confirmed

CAUTION: not confirmed on an official page. The official URL now serves the redesigned TC260 homepage, with no trace of the document, in all stored forms. Title, original title, identifier TC260-003, issuer, release date (29 Feb 2024), scope and the summary's content rest on CSET's unofficial English translation (secondary): its page and the full translation PDF are stored. No Chinese text is stored; the translation names the Chinese source as tc260.org.cn/upload/2024-03-01/1709282398070082466.pdf, and copies at perma.cc (HTTP 403, browser timeout) and c-csa.cn (robots.txt) could not be stored in the earlier run. The authors' own English title, as CSET reports it, renders 安全 as 'security', and the summary follows it; CSET's translation uses 'safety'. 'Superseded' is a judgment: no TC260 withdrawal notice was found, and GB/T 45654-2025 (CN-10) took up the content.

Second check: Independent check by Claude Code (claude-sonnet-5-5), 2026-10-14: 10 field(s) supported.

History

2026-10-14notes: Released 29 Feb 2024 and posted on the TC260 site 1 Mar 2024. Its content was ra → Released 29 Feb 2024; a preliminary draft had been published in October 2023. It — Removed 'Status Superseded is our assessment: we found no formal TC260 withdrawal notice' and the remark on the secondary URL (both now in verification.note), 'CSET notes that ...' (source wording), and the 1 Mar 2024 posting date (in no stored file). Added the October 2023 preliminary draft.2026-10-14-refresh-translation-fixes · Claude Code (claude-opus-5-5)
2026-10-14safety_topics: ["general-purpose / frontier AI", "evaluation & testing"] → ["general-purpose / frontier AI", "evaluation & testing", "catastrophic risk"] — Return round (14 Oct 2026): added 'catastrophic risk': section 4 asks providers to watch AI able to deceive humans, self-replicate or self-modify, and its use to create biological or chemical weapons.2026-10-14-refresh-translation-fixes · Claude Code (claude-opus-5-5)
2026-10-14summary: TC260 technical document that set baseline security requirements for generative → TC260 technical document setting the basic security requirements that generative — Return round (14 Oct 2026): rewritten from the full CSET translation, now stored. Restored, with evidence, the link to filing under the Interim Measures (section 4 and 9.1: assessment reports are submitted for filing) and the 31 risks of Appendix A (A.1 to A.5: 8 + 9 + 5 + 7 + 2). Added the 5% corpus threshold (5.1) and the long-term risks in section 4. 安全 stays 'security', as in the authors' English title. First pass: removed '31' and the filing link because no stored file then had them.2026-10-14-refresh-translation-fixes · Claude Code (claude-opus-5-5)
2026-10-14english_version: – → {"kind": "third-party translation", "url": "https://cset.georgetown.edu/publicat — English version found by the translation check in a stored copy (archive/CN-06/2026-10-14/secondary.html).manual-2026-10-14 · Claude Code (claude-opus-5-5)

Related records

GB/T 45654-2025 China · Published (non-binding, current)

Generative AI Measures China · In force

Citation

National Cybersecurity Standardization Technical Committee (TC260). Basic Security Requirements for Generative Artificial Intelligence Services (TC260 technical document) [TC260-003]. 2024-02-29. https://www.tc260.org.cn/portal/article/2/20240301164054. Status: Superseded; verified 2026-10-14 (level C). Local copy archive/CN-06/2026-10-14/official.html retrieved 2026-10-14 07:25:16 UTC, SHA-256 b882626d90f68a77283c19776a2fd8ca948768c4721810a260c81965c0a9aa59. Every AI Policy Ever, version 1.18 (snapshot 2026-10-18), record CN-06.

Improve this record

Know of an amendment, a status change or an error? Report news or an error on GitHub; the form opens with this record filled in.