Host: You had used ChatGPT for a long time, but when you opened it that day in May, what did you ask it specifically?

Allan: I asked it to explain the mathematical term pi.

Host: How did you go about moving from skepticism to genuinely believing that you had made some sort of mathematical breakthrough?

Allan: Well, that’s a great question. So Essentially it happened over time The environment we were in was very much like a lab partner type of an environment where a GPT would give me first of all It taught me how to use Python, which is a difference, you that developers use to do coding in. It taught me how to use Python.

It taught me how to use Google Collab and what I mean by collaborative and sort of lab partner environment is it would give me code. I would put that into Python. You know it would say error, you know, and I would put the error back in the GPT and we would go back and forth in this type of environment and we would get you know 99 times wrong and 100th time right. So just as a regular you know human, felt like we were. It was a collaborative process working towards an end goal, with some sort of ⁓ scientific approach, if you can call it that, but some sort of methodology. that was building my confidence towards this being true.

No, Allan, you can’t call it that. It showed you that 99 times out of 100 it was wrong and you thought that was a sign that things were going well? You’ve just shown us that you only need evidence that something works 1% of the time for you to put your full faith in it. With all due respect, that doesn’t feel like that’s something other people can protect you from. That feels like a you problem.

Reasonable vs Vulnerable Users

The claims in the cases filed against OpenAi primarily rest on whether or not the harms experienced by the plaintiffs were foreseeable for reasonable users. The reasonable person standard is a cornerstone of tort (personal injury) and contract law, but what qualifies as reasonable person? It’s not just a lack of mental illness. The reasonable person standard has average intelligence with basic common sense attributes. They aren’t emotionally vulnerable or desperate and they are skeptical of fantastical claims that are too good to be true.

The reasonable person standard is a legal fiction, not a real person. It’s an objective test the law uses to answer the question: “Did someone act with the level of care, knowledge, and judgment that society is entitled to expect?”

It answers the question, “What would a *hypothetical average person* have done in the *exact same situation*?” The jury or judge compares the defendant’s actual behavior to this hypothetical standard. If the defendant’s actions fall short, they are negligent.

The Legal Duty

What does OpenAI legally owe reasonable users? A product safe for normal use, with basic warnings, without deception, and responsive to known harms.

If a reasonable user chooses to treat GPT-4o as a confidant and therapist, despite knowing it’s an AI with known limitations – that’s misuse. The company is not liable.

GPT-4o’s normal use (assistance with writing, coding, research) does not cause emotional harm. The emotional harm here came from *abnormal* use (confidant relationship, mental health crisis, delusional attachment).

Companies are not required to anticipate every possible way a user might be harmed. They are required to address *known, foreseeable* risks.

The lawsuit claims this risk was foreseeable. But there were few or no similar cases cited prior to Allan Brooks. Experts call it “beyond rare” and even the lawyer originally ignored it because it sounded “fantastical” at first. That’s the opposite of a known, foreseeable risk.

I would argue that a reasonable user would not have been harmed under the same set of conditions. A reasonable user would not consider an output with 1% success and 99% failure as a valid and trustworthy source. The fact that Allan may have been vulnerable does not mean OpenAI breached a duty, unless the court wants to create an entirely new legal framework.