AI得管起来了！全球首部全面监管AI法律出台

The new rules ban certain AI applications that threaten citizens’ rights, including biometric categorisation systems based on sensitive characteristics, untargeted scraping of facial images from the internet or CCTV footage to create facial recognition databases, emotion recognition and AI that manipulates human behaviour.

新规定禁止某些威胁公民权利的应用，如通过记录特征进行生物识别分类的系统，非针对性的从互联网或闭路电视录像中获取面部图像并建立面部识别数据库的，识别情绪和操纵人类行为的AI。

点击阅读AI是如何教唆行刺的↓

执法豁免

Law enforcement exemptions

The use of biometric identification systems (RBI) by law enforcement is considered a high-risk use case, requiring judicial authorisation being linked to a criminal offence, which is prohibited in principle.

原则上禁止执法机构使用生物识别系统（RBI），这类系统被视为高风险类应用，执法时需要刑事犯罪相关的司法授权。

高风险系统承担义务

Obligations for high-risk systems

Clear obligations are also foreseen for other high-risk AI systems, include critical infrastructure, employment, migration, justice and democratic processes (e.g. influencing elections). Such systems must assess risks, maintain use logs, be transparent, and ensure human oversight. 

对于其他高风险人工智能系统也明确规定了义务，如应用于关键基础设施、就业、移民、司法和民主进程（例如影响选举）。这些系统必须接受风险评估、保留使用日志、保持透明，并确保人类监督。

Citizens will have a right to submit complaints about AI systems and receive explanations about decisions based on high-risk AI systems that affect their rights.

公民将有权对AI系统提出投诉，必须收到关于这种影响公民权利的AI高风险系统使用的合理解释。

透明度要求

Transparency requirements

It also requires all AI-generated deepfakes to be clearly labeled, targeting concerns about manipulated media that could lead to disinformation and election meddling.

法案还要求必须明确标注所有AI生成的深度伪造（deepfakes），回应了对可能存在的虚假信息和可能导致的选举干扰的担忧。

重点监管对象：生成式AI

Generative Pre- trained Transformer

There’s extra scrutiny for the biggest and most powerful AI models that pose “systemic risks,” which include OpenAI’s GPT4 — its most advanced system — and Google’s Gemini.

对于可能带来 "系统性风险"的最大型、最强的人工智能模型，包括目前最先进的 OpenAI 的 GPT4 和谷歌的 Gemini，法案加强了特殊监管。

Companies that provide these systems will have to assess and mitigate the risks; report any serious incidents, such as malfunctions that cause someone’s death or serious harm to health or property; put cybersecurity measures in place; and disclose how much energy their models use.。

法案规定，提供这些系统的公司必须评估和降低风险；报告任何严重事故，如导致人员死亡或严重损害健康或财产的故障；采取网络安全措施；并披露其模型的能源使用情况。

The EU says it’s worried that these powerful AI systems could “cause serious accidents or be misused for far-reaching cyberattacks.” They also fear generative AI could spread “harmful biases” across many applications, affecting many people.

欧盟担心，这些强大的人工智能系统可能 “导致严重事故或被滥用于影响深远的网络攻击”。而且生成式人工智能可能会在许多应用中传播“有害的偏见”，影响许多人。

Violations of the AI Act could draw fines of up to 35 million euros ($38 million), or 7% of a company’s global revenue.

违反《人工智能法》最高可被罚款 3500 万欧元（约合人民币 2.74 亿元），或公司全球收入的 7%。

The legislation is the result of a proposal that was first introduced in 2021, when the release of ChatGPT spurred an investment boom and public frenzy.

该立法是 2021 年首次提出的一项提案的后续结果，当时 ChatGPT 的问世引发了投资热潮和公众狂欢。

The regulation, agreed in negotiations with member states in December 2023, was endorsed by MEPs with 523 votes in favour, 46 against and 49 abstentions.

该法规于 2023 年 12 月与成员国谈判达成一致，以 523 票赞成、46 票反对和 49 票弃权获得欧洲议会议员的认可。

Frédéric Marvaux/European Union

而世界其他国家，虽然出台了一些AI相关的法律，但都止步于单一条例。

2023 年 7 月，中国网信办发布《生成式人工智能管理暂行办法》，从 8 月 15 日开始实施。《办法》提出，国家采取有效措施鼓励生成式人工智能创新发展，对生成式人工智能服务实行包容审慎和分类分级监管。

In October 2023, US President Joe Biden announced an executive order requiring AI developers to share data with the government.

2023 年 10 月，美国总统拜登在一项行政命令中要求人工智能开发人员与政府共享数据。

In November 2023, the UK hosted an AI safety summit but is not planning legislation along the lines of the AI Act.

2023 年 11 月，英国主办了一次人工智能安全峰会，但没有按照《人工智能法案》的思路计划立法。

"(This is) the first regulation in the world that is putting a clear path towards a safe and human-centric development of AI," Benifei said.

“（这是）世界上第一个考虑安全和以人为本的AI发展开辟明确道路的法规，” 意大利的欧洲议会议员白兰度·贝尼菲（Brando Benifei）表示。

Getty images

The regulation is still subject to a final lawyer-linguist check. The EU's 27 states are expected to endorse the text in April before the law is published in the EU's Official Journal in May or June.The AI Act still needs to be endorsed by individual EU member states, a step expected to happen in April.The legislation would then come into force later this year.

该法规仍需经过律师和语言学家的最终检查。欧盟 27 个国家预计将在 4 月批准该法律，然后在 5 月或 6 月在欧盟官方公报上公布。该立法将于今年晚些时候生效。

图：欧盟立法流程

Rules covering generative AI models such as ChatGPT will come into effect 12 months after the law becomes official, while companies must comply with most other provisions within two years. Prohibited AI systems will be banned six months after the law comes into effect.

涵盖 ChatGPT 等生成式 AI 模型的条例将在法律正式生效 12 个月后生效，而公司必须在两年内遵守大多数其他规定。被禁止的 AI 系统将有六个月的缓冲时间。