社團指南

人工智慧造成傷害時誰該負責?

When AI causes harm, liability usually falls on people and companies rather than on the AI.

聆聽此指南AI 語音敘述 · ~3 min

本頁4 分鐘閱讀
  1. 概述
  2. 深入探討
  3. 現實世界的實施
  4. The Future of Who Is Liable When AI Causes Harm?
  5. 付諸實踐
  6. 不斷探索
  7. 常見問題

概述

That can be the business that deployed it, the developer or manufacturer of a defective product, or a professional who relied on it carelessly, and the claim is usually brought under existing law such as negligence, misrepresentation and product liability. It matters because courts and lawmakers are now deciding how these old rules apply to software that learns and changes. In the EU, product liability rules have been rewritten to cover software, including AI.

深入探討

Most AI harm claims fit into three familiar legal routes. Negligence asks whether someone failed to take reasonable care, for example by deploying a system without adequate testing or monitoring. Misrepresentation covers false statements a business makes, including statements made through its chatbot. Product liability can hold manufacturers strictly liable for defective products, meaning the injured person does not have to prove fault, only the defect, the damage and the causal link. The Air Canada case shows the simplest principle. In February 2024, the Civil Resolution Tribunal found the airline liable for negligent misrepresentation after its chatbot wrongly said bereavement fares could be claimed retroactively. The award was small, roughly C$800 including interest and fees, but the reasoning is clear: a company is responsible for all the information on its website, whether it comes from a static page or a chatbot. In the EU, the revised Product Liability Directive (EU) 2024/2853 entered into force in December 2024 and applies to products placed on the market from December 2026. It explicitly treats software, including AI systems, as a product. Free and open-source software developed outside commercial activity is excluded. Covered damage now includes medically recognized psychological harm and the destruction or corruption of data not used for professional purposes. Courts can order the disclosure of evidence and presume defect or causation where technical complexity makes proof excessively difficult. The separate AI Liability Directive, proposed in 2022 to ease fault-based claims, was dropped after the Commission announced in February 2025 that it expected no agreement. A common misconception is that AI creates a legal vacuum. In practice, existing doctrines are applied, although proving causation for opaque systems remains hard.

現實世界的實施

In Moffatt v. Air Canada (2024), a British Columbia tribunal held the airline responsible after its website chatbot gave wrong information about bereavement fares. It rejected the argument that the chatbot was responsible for its own statements.

In Mata v. Avianca (2023), lawyers in New York were sanctioned for filing a brief with fake case citations produced by ChatGPT. The court held the professionals responsible for checking their filings.

Under the revised EU Product Liability Directive, a person injured by a defective AI-enabled medical device or robot can claim compensation from the manufacturer without proving fault.

In Garcia v. Character Technologies, a family alleged that a chatbot's design contributed to a teenager's death. In 2025 a US federal judge let product liability claims proceed past an early motion to dismiss, testing whether chatbot apps can be treated as products.

The Future of Who Is Liable When AI Causes Harm?

Expect more litigation that tests whether generative AI outputs count as products, how much protection the US Section 230 gives AI-generated content, and how courts handle causation for opaque systems. In the EU, the revised Product Liability Directive will start applying to new products from late 2026, and national courts will begin interpreting its presumptions. The withdrawn AI Liability Directive leaves fault-based claims to national law for now, so results may vary between countries. For organizations, careful documentation and human oversight are likely to matter more than any single new law.

付諸實踐

為什麼這很重要

  • 風險與安全. 災難性和日常的人工智慧危害都取決於誰了解風險以及誰能夠採取行動。
  • 更明確的決策. 民眾和專業素養決定強而有力的安全政策在政治上是否可行。
  • 突破炒作. 清晰的解釋可以減少炒作、實驗室公關和模糊道德劇場的影響。

留意

  • 將存在風險視為科幻小說,同時能力複合。
  • 混淆了表面產品安全與高度自治下的對準。
  • 只給非英語和非專業觀眾留下低品質的資源。

後續步驟

  1. 單獨的產品危害、誤用和失控/失調風險。
  2. 詢問哪些證據會改變您對時間表和嚴重性的看法。
  3. 比起行銷主張,更喜歡主要來源和具體評估。
  4. 確定一條行動路徑:職業、政策、資金或技能——而不僅僅是意識。

不斷探索

免費電子報

獲取每日 AI 簡報

每個工作日上午提供三則經過驗證的 AI 新聞,以淺顯易懂的英文撰寫。永久免費,無廣告。

每個工作日一封電子郵件。一鍵取消訂閱。我們絕不出售或分享您的地址。

測試一下自己

Take the Who Is Liable When AI Causes Harm? quiz

每個答案皆有即時回饋,通過課程後可獲得具備可驗證 ID 的可分享證書。

開始測驗

支持免費的 AI 教育。AI Understanding 是一個 501(c)(3) 非營利組織——絕無廣告,絕無付費牆。 捐款

常見問題

人工智慧造成傷害時誰該負責?

When AI causes harm, liability usually falls on people and companies rather than on the AI. That can be the business that deployed it, the developer or manufacturer of a defective product, or a professional who relied on it carelessly, and the claim is usually brought under existing law such as negligence, misrepresentation and product liability. It matters because courts and lawmakers are now deciding how these old rules apply to software that learns and changes. In the EU, product liability rules have been rewritten to cover software, including AI.

法庭在莫法特诉加拿大航空案中做出了什么裁决?

仲裁庭发现存在疏忽失实陈述,并认为公司应对其网站上的所有信息负责,包括聊天机器人的答案。

严格的产品责任是什么意思?

严格责任消除了证明粗心的必要,这使得受伤者更容易获得赔偿。

修订后的欧盟产品责任指令如何对待软件?

指令 (EU) 2024/2853 将软件纳入产品责任范围。在商业活动之外开发的免费和开源软件不包括在内。

What happened to the proposed EU AI Liability Directive?

The Commission abandoned the proposal, which leaves fault-based AI claims to national laws for now.

Which new type of damage does the revised EU directive cover?

Covered damage now includes loss or corruption of data not used for professional purposes, as well as medically recognized psychological harm.