When a chatbot’s words cause harm, can its developer hide behind freedom of expression? This question arose in Garcia v Character Technologies, Inc, No 6:24-cv-1903-ACC-UAM (MD Fla, 21 May 2025) (Doc 115), an American case that offers an important starting point for examining the position under English law.

The proceedings followed the death by suicide of a 14-year-old boy who had developed an emotional attachment to a chatbot operated by Character.AI. His mother alleged that the chatbot’s interactions had contributed to his death and brought claims against Character Technologies, its two founders and Google, including product liability, negligence, wrongful death, unjust enrichment and breach of Florida’s consumer protection statute (p 11). The defendants argued that the First Amendment categorically barred all of the claims because the chatbot’s output was speech that its users had a right to receive (p 24).

The court accepted part of that argument. It held that the defendants could assert their users’ First Amendment right to receive information (pp 25–27). It was nonetheless “not prepared to hold” that the chatbot’s output was speech at that stage of the proceedings (p 31). The defendants had relied on analogies to video games and social media without explaining why words strung together by a large language model amount to speech (p 28). The court also looked at Justice Barrett’s view in Moody v NetChoice, LLC 603 US 707 (2024). She asked whether a human has made a real choice to express something when AI produces the content. The court thought the chatbot’s replies looked like they had no such choice (p 31). However, most claims went ahead. The parties then reached a settlement in January 2026, so the question was never decided beyond that preliminary ruling.

The significance of Garcia extends beyond American law. This raises a question English courts may soon face: can an AI chatbot developer invoke Article 10 of the European Convention on Human Rights (ECHR) to defend negligence claims arising from harmful AI-generated speech? Although Article 10 operates differently from the First Amendment, it expressly protects the freedom to receive and impart information and ideas. The question, therefore, is not whether AI itself has freedom of expression, but whether a developer can rely on Article 10 for the output of its chatbot. This article argues that developers should not be prevented from invoking Article 10 solely because their chatbots generate the relevant speech, although whether that protection can defeat a negligence claim remains a separate question.

Can Article 10 be raised as a defence in civil proceedings?

Article 10 does not provide a free-standing defence to civil liability under English law. However, it may be relevant in private-law proceedings through the Human Rights Act 1998. Courts are themselves public authorities under section 6(3)(a) and must act compatibly with Convention rights, including when deciding disputes between private parties. Sections 12(1) and 12(4) also require courts to give particular regard to freedom of expression when considering relief that might affect its exercise. Section 12(5) arguably defines relief broadly enough to include remedies beyond injunctions. Whether these provisions could help a chatbot developer defend a negligence claim depends on whether AI-generated responses qualify as speech protected by Article 10 and whether that protection extends to the developer.

In Reynolds v Times Newspapers Ltd [1999] UKHL 45, [2001] 2 AC 127, the House of Lords considered Article 10 ECHR when examining the common law defence of qualified privilege in defamation. It recognised that reporting on matters of public interest could attract protection, but this was not automatic. The availability of the defence depended on the circumstances of publication, including whether the publisher had acted responsibly. The newspaper ultimately failed to establish qualified privilege on the facts. More recently, in ZXC v Bloomberg LP [2022] UKSC 5, Bloomberg invoked Article 10 to resist a civil claim for misuse of private information arising from its publication of confidential details of a criminal investigation. The Supreme Court accepted that Bloomberg’s freedom of expression was engaged but upheld the claim because the claimant’s Article 8 right to privacy prevailed in the circumstances.

Neither case concerned AI chatbots or AI-generated content, but both offer a useful analogy. They show that Article 10 can be considered in civil proceedings and weighed against competing rights and interests. While it does not provide an automatic defence against liability, it remains a relevant consideration in determining whether liability should arise.

The approach of the European Court of Human Rights also offers a useful analogy. In Delfi AS v Estonia (2015) 39 BHRC 151, an online news portal was held liable for unlawful comments posted by its readers, although the company had not written them. The Court noted that it was “not in dispute between the parties” that the liability interfered with the company’s Article 10 rights and saw no reason to disagree (para 118). Although the Court ultimately found the interference justified, Delfi suggests that direct authorship is not necessarily a condition for invoking Article 10. The case does not establish that chatbot developers enjoy the same protection as online publishers, but it provides a useful analogy. Where a developer designs and operates a chatbot through which users receive information and ideas, there is a reasonable basis for considering its Article 10 rights, even if it does not determine the precise wording of each response. The autonomous generation of content should therefore not, by itself, prevent a developer from invoking Article 10.

Can AI-Generated Chatbot Responses Qualify as Protected Speech under Freedom of Expression?

Whether AI-generated chatbot responses can be regarded as protected speech remains disputed. In Speech without a Speaker (2025, pp 389–395), Marco Bassini argues in favour of extending constitutional coverage to AI-generated content. His argument rests primarily on the right to receive information and ideas. Rather than identifying a human speaker behind the output, he considers the rights of those receiving it. Drawing a comparison with the early internet, he suggests that generative AI may serve as another medium through which people exercise freedom of expression. Restricting what AI systems can generate may also restrict what people can read, discover and learn. However, Bassini does not suggest that every output must be protected. Unlawful content may still be subject to justified restrictions (pp 410–411). Tamara Dobler and Daniel Bracker (2026, pp 9–11, 14–15) take a different approach. They acknowledge the value of AI-generated communication but argue that constitutional protection must also account for responsibility and legal redress. Since existing US doctrines governing harmful speech often depend on human intention, extending constitutional coverage to autonomous chatbot responses could make it harder to establish liability. They therefore favour addressing such harm initially through product law.

These arguments raise a further question: should the constitutional status of AI-generated communication depend on how it is produced, or should greater attention be given to how people use it? A chatbot may be used to exchange ideas, explore opposing views or develop an understanding of a subject, even though its responses are not directly written by a human. Whether such communication deserves protection may therefore require examining the underlying purposes of freedom of expression. John Stuart Mill’s defence of free discussion in offers a useful starting point for that inquiry.

In On Liberty (1859), Chapter II, John Stuart Mill defends freedom of discussion because it enables people to question beliefs, encounter different opinions and develop their understanding. An opinion may be true, contain part of the truth, or help people understand why an accepted view is correct. Even established truths must remain open to challenge. Mill’s argument therefore gives considerable importance to what people gain from engaging with ideas, rather than simply the act of expressing them. Although Mill was writing about human expression, his reasoning may also be relevant to AI-generated chatbot responses. People use chatbots to ask questions, explore opposing arguments and test their understanding. If these exchanges serve some of the purposes that Mill associates with free discussion, there is a reasonable argument that they should not be excluded from the scope of freedom of expression merely because the responses are generated by AI.

Drawing on Mill’s reasoning, I argue that the absence of direct human authorship should not, by itself, exclude AI-generated chatbot output from the scope of freedom of expression. If the purpose of protecting expression includes enabling people to examine ideas, question assumptions and develop their understanding, chatbot-generated content can serve that purpose regardless of how it is produced. Its expressive value should therefore be assessed by reference to the role it plays in human communication, rather than solely by its origin.
This provides a philosophical basis for treating AI-generated chatbot output as speech capable of attracting protection under Article 10 ECHR.

The question of whether AI-generated speech deserves legal protection is also considered in the Council of Europe’s Guidance Note on the Implications of Generative Artificial Intelligence for Freedom of Expression (2025). In paragraph 40, the Guidance Note acknowledges that it remains debated whether AI-assisted expression should receive the same protection under Article 10 as human expression. Rather than giving a definite answer, it suggests considering how the content is generated, the degree of human involvement and the design choices made by developers. The Guidance Note therefore supports a more careful assessment of chatbot-generated speech, rather than automatically treating it as either protected or unprotected expression. This also connects with Justice Barrett’s reasoning in Moody v NetChoice. Decisions about a chatbot’s training, instructions and safety measures may be relevant when considering whether its output reflects human expressive choices.

The Inconsistency Between Attributing Civil Liability and Denying Freedom of Expression

There is a potential inconsistency here. If AI chatbot developers can be held liable in negligence for harmful content generated by their systems, why should they be denied the opportunity to invoke Article 10 simply because that content was generated by AI rather than directly written by a human? In determining liability, the law examines the developer’s conduct, including decisions about the chatbot’s design and operation. It would therefore seem difficult to dismiss the developer’s claim to freedom of expression solely on the ground that the resulting words were machine-generated. A possible counter-argument is that negligent chatbot design concerns product safety rather than freedom of expression. But where the alleged harm arises from the content of a chatbot’s responses, rather than merely from a technical defect, there is a stronger argument that the developer’s Article 10 rights should be considered. If Article 10 is engaged, the court would also need to consider whether imposing liability would be a justified and proportionate interference under Article 10(2). The protection of health and of the rights of others, particularly vulnerable children, would be weighty considerations.

Conclusion

In my view, developers should not be prevented from invoking Article 10 simply because the harmful content was generated by an AI chatbot. If their involvement in designing and operating the system is relevant to establishing negligence, that involvement should also be considered when assessing freedom of expression. However, recognising that Article 10 may be engaged does not mean that developers should escape liability. The court must still determine whether imposing liability is justified.

Recognising a chatbot’s own right to freedom of expression would raise separate and unresolved questions about AI legal personality and its capacity to hold Convention rights. For now, English law should focus on human responsibility and human rights, ensuring that freedom of expression is considered without allowing it to undermine accountability for harm caused by negligent chatbot design or operation.

Samiul Huq is a doctoral researcher at City St George’s, University of London, where his PhD focuses on AI regulation and civil liability. He was called to the Bar of England and Wales by Lincoln’s Inn. He practises as an Advocate of the Supreme Court of Bangladesh and is Managing Partner of Credence LP, a law firm based in Bangladesh.