Is AI Responsible for Misleading Us?

A recent court case in Germany found Google liable for false statements that were generated by its AI overview. Two publishers discovered that when users searched the names of their companies, Google’s AI presented information confidently stating the companies were scams with a history of fraud (which they weren’t). Google, while losing the case, did not go down without a fight (and is already planning to appeal the decision). In their defense, they argued that the warning at the bottom of their search results – which reads, at the time of writing, “AI can make mistakes, so double-check responses” – is sufficient to free them from any liability related to anything in the AI summary itself. According to the company, those who rely on Google’s AI summaries should know “that information generated with AI should not be blindly trusted,” and therefore it is the responsibility of users, not Google, to ensure that what they are reading is true.

This argument did not fly in court. But Google’s defense underscores the sometimes-contradictory ways tech companies have attempted to avoid accountability when it comes to the creations of their chatbots. Specifically, tech companies will selectively argue that their AIs are or are not responsible for their outputs, depending on what best suits their needs.

Whether Google should be held responsible for its AI’s behavior, despite providing a warning that it could be providing users with false information, depends in part on how we conceive of the chatbot itself. Consider an example: let’s say I ask my friend for directions to a restaurant he’s recommended (also say the restaurant is brand new and so doesn’t show up on Google Maps yet). He confidently gives me the directions, but adds a caveat: “I’m only human, and none of us are perfect.” Not taking too much heed of his odd remark, I start on my way and quickly find myself lost: it turns out my friend’s directions were completely wrong. When I ask him what happened, he replies that he’s not responsible for my mistake and that I should have double-checked. After all, I really should know better than to blindly trust other people, and he did remind me he can be wrong sometimes.

It seems my friend cannot get off the hook so easily: he told me something that turned out to be false, and for that, he at least owes me an apology (and maybe dinner). Merely reminding someone that there’s always a chance that we could be wrong doesn’t relieve us of our responsibility: if we were really that concerned about saying something false, then we’d either have to retract a statement if we made it or not make it at all.

Of course, this is the kind of responsibility that people have, and AIs are not people. People have agency, and thus can make intentional choices about what they say. But an AI is different. Indeed, Google argued in the German court that their AI merely conveys information that other people have written. In this way, Google argued that its chatbot should be held to the same standards as your run-of-the-mill search engine: since Google search results just present links to stuff that other people have made, Google is not responsible for what shows up in its results (by and large). Similarly, Google argued that in virtue of their AI presenting other people’s content and not its own, they bear no responsibility for it.

As we’ve seen, this argument did not save Google’s case in the German court. One reason the court ruled against Google is that the AI summary hallucinated new information that could only be attributed to the program itself, and thus the responsibility for the defamatory content it presented fell to Google. However, Google’s overall strategy was as follows: argue that its AI should not be compared to the kinds of things that are responsible for their outputs. In many other cases, however, tech companies have argued the exact opposite to try to get out of trouble.

The case in Germany is by no means the first time tech companies have gotten into hot water for their chatbots causing harm to users. For example, in 2025, a court case in the US was brought against tech company Character Technologies when its AI was alleged to have played a large part in a teen’s suicide. In response, the company argued they should not be held liable since their chatbot’s outputs constituted free speech. This argument was rejected by the court on the basis that chatbots don’t produce “speech” – they merely present the outcomes of algorithms – and thus could not benefit from the protections of free speech.

The strategy that Character Technologies took, however, was the opposite of Google’s, arguing that the AI program should be treated in the same way as things that are responsible for their outputs so that they are granted the rights that accompany those responsibilities. While this would not save the company from any moral responsibility, possessing the right of free speech could potentially save them from legal liability.

A similar argument can also be found in ongoing lawsuits concerning the use of copyrighted material to train AI models. As I’ve written about here, some tech companies have argued that training large language models using copyrighted materials is “fair use,” given that using such works in this way is “transformative,” and that training an AI model is akin to the ways that human beings learn and produce new content. Since there’s nothing wrong with a person reading a lot of copyrighted books and then producing new content on the basis of what they have read, so goes the argument, then neither should there be any problems with an AI doing the same thing.

This argument has again been rejected by courts (at least thus far): while a person has the capacities to learn, be creative, and produce something new, and thus also has the responsibility to use content fairly, a computer program that produces content on the basis of algorithms does not have any of these capacities or responsibilities and so lacks any accompanying rights.

Tech companies, then, face a dilemma when trying to skirt moral and legal responsibilities when their chatbots produce false, harmful, or defamatory content: either they are treated like things that can be responsible for their actions, or they are not. If the former, then perhaps they should be granted certain rights, but then they cannot simply be treated as programs that reproduce the content of others, and so their creators must bear responsibility for their content. If the latter, then a case could be made that the responsibility for the content they produce should fall to the original creators of that content that they are reproducing, but then the tech companies must admit that they are, in fact, not using other people’s creations in a fair or transformative way, and so bear responsibilities towards those creators (namely, to pay them).

While there are many court cases around AI that still need to be resolved, appealed, re-appealed, and so on, it seems that no matter what, tech companies cannot completely avoid liability when it comes to content their chatbots create.