全国网络安全标准化技术委员会 2024年9月 人工智能 安全治理框架1. 人工智能安全治理原则 …………………………………… 1 2. 人工智能安全治理框架构成 ……………………………… 2 3. 人工智能安全风险分类 …………………………………… 3 3.1 人工智能内生安全风险 ……………………………… 3 3.2 人工智能应用安全风险 ……………………………… 5 4. 技术应对措施 针对人工智能内生安全风险 ………………………… 7 4.2 针对人工智能应用安全风险 ………………………… 9 5. 综合治理措施 ……………………………………………… 10 6. 人工智能安全开发应用指引 ……………………………… 12 6.1 模型算法研发者安全开发指引 ……………………… 12 6.2 人工智能服务提供者安全指引 ……………………… 13 6.3 重点领域使用者安全应用指引 6.4 社会公众安全应用指引 ……………………………… 15 目 录- 1 - 人工智能安全治理框架 人工智能是人类发展新领域，给世界带来巨大机遇，也带来各类风险挑战。 落实《全球人工智能治理倡议》，遵循“以人为本、智能向善”的发展方向，为 推动政府、国际组织、企业、科研院所、民间机构和社会公众等各方，就人工 智能安全治理达成共识、协调一致，有效防范化解人工智能安全风险，制定本 框架。

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465 20.1. 不安全 Rust . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 昭著的陷阱。即使谨慎的实践者，亦唯恐代码出现漏洞、崩溃或损坏。 Rust 破除了这些障碍：它消除了旧的陷阱，并提供了伴你一路同行的友好、精良的工具。想 要 “深入” 底层控制的程序员可以使用 Rust，无需时刻担心出现崩溃或安全漏洞，也无需因为 工具链不靠谱而被迫去了解其中的细节。更妙的是，语言设计本身会自然而然地引导你编写出 可靠的代码，并且运行速度和内存使用上都十分高效。 已经在从事编写底层代码的程序员可以使用 Rust abstractions）—— 将高级语言特性编 译成底层代码，并且与手写的代码运行速度同样快。Rust 努力确保代码又安全又快速。 这里提到的只是几个较大的受益群体，Rust 语言也希望能支持更多其他用户。总的来说， Rust 最重要的目标是消除数十年来程序员习以为常的取舍，让安全和高效、速度和易读易用 可以兼得。试试看 Rust，说不定它的选择就适合你。 本书适合哪些人 本书假设你已经有其他

面对全球大模型产业之争，要打赢「三大战役」 AGI之战 应用场景之战 大模型安全之战 • 探索超越人类的超级人工 智能AGI • 不仅是科技之争，更是国 运之争 • 不发展是最大的不安全， 发挥举国体制优势，打赢 追赶之战 • 大模型带来前所未有安全 挑战 • 外挂式传统安全手段难以 应对 • 应对模型安全新挑战，打 赢未雨绸缪之战 • 大模型是能力而非产品， 结合场景才能发挥价值 创业公司得到DeepSeek加持，创业者拥有便宜领先的大模型，迎来 机遇，带来“iPhone时刻” 中国变成AI渗透率最高的国家，率先实现AI工业革命 37政企、创业者必读 人人智能 万物智能 数转智改 未来产业 科学研究 安全 应用爆发的六大方向 38政企、创业者必读 DeepSeek的开源和低成本使得个人也能够拥有自有大模型，实现超能力， 成长为超级个体 DeepSeek六大应用方向之一 人人智能：人人都要用AI 从数年缩短到几分钟，解开了生物学密码 成功预测了地球存在的2亿种蛋白质结构 45政企、创业者必读  DeepSeek典型的四大安全问题：客户端安全、Agent安全、知识安全、模型安全  360提出「以模制模」新解法，应对DeepSeek安全问题 DeepSeek六大应用方向之六 AI安全：实现安全的「自动驾驶」 46政企、创业者必读 大模型的六大能力 47 基本 能力 业务 能力 创新 能力

ERROR: UndefRefError: access to undefined reference This avoids the need to continually check for null values. However, not all object fields are references. Julia considers some types to be "plain data" raw pointer value must be called at runtime for precompilation to work (Ptr objects will turn into null pointers unless they are hidden inside an isbits object). This includes the return values of the the missing object, which is the singleton instance of the type Missing. missing is equivalent to NULL in SQL and NA in R, and behaves like them in most situations. 21.1 Propagation of Missing Values

ERROR: UndefRefError: access to undefined reference This avoids the need to continually check for null values. However, not all object fields are references. Julia considers some types to be "plain data" raw pointer value must be called at runtime for precompilation to work (Ptr objects will turn into null pointers unless they are hidden inside an isbits object). This includes the return values of the the missing object, which is the singleton instance of the type Missing. missing is equivalent to NULL in SQL and NA in R, and behaves like them in most situations. 21.1 Propagation of Missing Values

ERROR: UndefRefError: access to undefined reference This avoids the need to continually check for null values. However, not all object fields are references. Julia considers some types to be "plain data" raw pointer value must be called at runtime for precompilation to work (Ptr objects will turn into null pointers unless they are hidden inside an isbits object). This includes the return values of the the missing object, which is the singleton instance of the type Missing. missing is equivalent to NULL in SQL and NA in R, and behaves like them in most situations. 21.1 Propagation of Missing Values

ERROR: UndefRefError: access to undefined reference This avoids the need to continually check for null values. However, not all object fields are references. Julia considers some types to be "plain data" raw pointer value must be called at runtime for precompilation to work (Ptr objects will turn into null pointers unless they are hidden inside an isbits object). This includes the return values of the the missing object, which is the singleton instance of the type Missing. missing is equivalent to NULL in SQL and NA in R, and behaves like them in most situations. 21.1 Propagation of Missing Values

ERROR: UndefRefError: access to undefined reference This avoids the need to continually check for null values. However, not all object fields are references. Julia considers some types to be "plain data" raw pointer value must be called at runtime for precompilation to work (Ptr objects will turn into null pointers unless they are hidden inside an isbits object). This includes the return values of the the missing object, which is the singleton instance of the type Missing. missing is equivalent to NULL in SQL and NA in R, and behaves like them in most situations. 21.1 Propagation of Missing Values

ERROR: UndefRefError: access to undefined reference This avoids the need to continually check for null values. However, not all object fields are references. Julia considers some types to be "plain data" raw pointer value must be called at runtime for precompilation to work (Ptr objects will turn into null pointers unless they are hidden inside an isbits object). This includes the return values of the the missing object, which is the singleton instance of the type Missing. missing is equivalent to NULL in SQL and NA in R, and behaves like them in most situations. 21.1 Propagation of Missing Values