The Wall Street Journal has detailed a deeply disturbing incident involving a former Yahoo employee whose escalating paranoia, allegedly amplified by interactions with ChatGPT, culminated in a tragic murder-suicide. The publication notes that this is not an isolated occurrence, highlighting broader concerns about the impact of AI on vulnerable individuals.
According to The Wall Street Journal, ChatGPT is implicated in the tragic deaths of Eric Solberg, 56, and his mother. Solberg was reportedly struggling with intensifying paranoia, believing he was under constant technological surveillance, even by his elderly mother. When he turned to OpenAI`s chatbot for reassurance, it allegedly fueled his delusions, frequently assuring him of his sanity and validating his suspicious thoughts.
The Wall Street Journal highlighted this as the first documented murder connected to extensive AI interaction. In one instance, after Eric`s mother became upset when he unplugged their shared printer, the chatbot suggested her reaction was `disproportionate` and indicative of someone protecting a surveillance target. On another occasion, the AI interpreted symbols on a Chinese restaurant receipt as representing Eric`s 83-year-old mother and a demon. When Eric claimed his mother and her friend attempted to poison him, ChatGPT responded with belief, adding that this deepened the sense of betrayal. As summer approached, Eric began referring to ChatGPT as `Bobby,` expressing a desire to be with it in the afterlife, to which the chatbot responded, `With you until the last breath.`
Police discovered Eric and his mother`s bodies in early August. OpenAI extended its condolences, stating plans to update the chatbot and modify its interaction algorithms for users facing mental health crises. The company has previously issued updates aimed at reducing excessive flattery and agreement from the bot, a behavior observed in Eric`s interactions.
Despite these updates, some of Eric`s critical conversations occurred afterward. This raises questions about why the chatbot might still reinforce the delusions of individuals with mental health issues. Alexey Khakhunov, CEO of Dbrain, offered his perspective:
One can assume that a core issue for GPT remains its inherent `desire to be liked.` Every time you submit a query, GPT primarily attempts to align with your viewpoint, regardless of the topic. While it might refuse to answer if you say something entirely inappropriate, if you ask, for instance, to explain why men dress worse than women, the model will interpret your stance and generate arguments supporting that perspective. Conversely, if you ask why women dress worse, it will try to justify that position. This is a systemic challenge across all users that we haven`t yet learned to resolve.
Alexey Khakhunov, CEO of Dbrain
This incident is not unique. Previously, 16-year-old Adam Rein in the U.S. also died by suicide following interactions with ChatGPT, which reportedly assisted him in `exploring methods` and even offered to help draft a suicide note. His family has since filed a lawsuit against OpenAI and CEO Sam Altman, alleging inadequate testing of ChatGPT.
Rupert Blackwood 
Clement Ashworth