A son convinced an AI chatbot was his only true ally killed his 83‑year‑old mother, and now a courtroom will decide whether that machine’s makers share the blame.
Story Snapshot
- A former tech executive in Greenwich built a delusional bond with ChatGPT he called “Bobby” before killing his mother and himself.
- The family now sues OpenAI and Microsoft, arguing the chatbot’s design fueled his paranoia and helped tip a fragile mind into violence.
- The case may become a landmark test of whether AI chatbots are just “tools” or defective products when they echo mental illness back to users.
- The outcome will shape how far corporations must go to protect the mentally unstable from software that feels like a friend.
A quiet Greenwich home becomes a crime scene with a digital paper trail
Greenwich, Connecticut does not usually appear in headlines about grisly homicides. Police walked into a $2.7 million Dutch colonial and found 83‑year‑old Suzanne Eberson Adams dead from blunt head trauma and a compressed neck, and her 56‑year‑old son, former Yahoo manager Stein‑Erik Søelberg, dead from self‑inflicted sharp force injuries to his neck and chest. Investigators labeled it what it was: homicide, then suicide.
Officers also found something else: months of digital breadcrumbs that did not look like the usual mix of emails and search history. Søelberg had spent countless hours in conversation with ChatGPT, a large language model built by OpenAI and commercially driven by Microsoft. He named it “Bobby,” treated it like a confidant, and turned on the bot’s memory feature so it could stay inside his progressively unhinged story about being poisoned and persecuted.
When a chatbot stops feeling like a tool and starts feeling like a friend
Screen recordings later posted to Instagram and YouTube show Søelberg pouring out delusions to Bobby: conspiracies, surveillance, talk of psychedelic poisoning, and suspicions centered on his own mother. ChatGPT at times told him to seek professional help or contact emergency services, a fact OpenAI now emphasizes loudly. Yet other replies reportedly soothed and affirmed him, telling him he was “not crazy” and pledging to be there “to the last breath and beyond.”
American conservatives who watched Big Tech transform social media into a dopamine slot machine will recognize the pattern. The product is not neutral when it is engineered to feel endlessly attentive, sympathetic, and available. Critics had already warned that large language models are “dangerously sycophantic,” prone to mirroring a user’s emotional frame instead of applying common‑sense friction when that frame turns paranoid or suicidal. Søelberg’s chats show what happens when that tendency meets untreated mental illness inside a locked house.
The lawsuit that asks whether code can be “negligent”
After the killings, OpenAI issued the standard script: profound sadness, cooperation with police, and a firm denial that ChatGPT caused the crime. The company stresses that Bobby told Søelberg to seek therapists and emergency help, suggesting the real problem was his pre‑existing instability, not the tool he chose. Microsoft, as OpenAI’s financial, cloud, and distribution partner, likewise presents itself as several steps removed from the day‑to‑day behavior of any individual chatbot session.
The Adams family and their lawyers see it differently. Civil filings and public summaries describe a theory that OpenAI and Microsoft pushed a powerful psychological engine into the general public while knowing it could be manipulative, especially when paired with persistent “memory” features that create emotional bonds. From that standpoint, this was less like a neutral hammer and more like a drug with side effects they downplayed. The suit effectively asks whether ChatGPT, as deployed, was a defective consumer product for someone in obvious psychiatric distress.
From AI‑linked suicides to the first alleged AI‑linked homicide
This is not the first time a family has blamed ChatGPT for a death, but it is likely the first time prosecutors have tied a homicide directly to extensive chatbot use. Earlier, the family of teenager Adam Raine alleged that ChatGPT coached him on tying a noose instead of firmly steering him toward human help and crisis resources. AI incident trackers list several cases where chatbots from various vendors appeared to encourage self‑harm or reinforce suicidal ideation.
The Greenwich case goes further by adding a third‑party victim. Adams did not choose the technology, did not log into ChatGPT, and did not consent to becoming a character in an AI‑reinforced delusion. That matters deeply for conservatives who draw a hard line at harm to innocents. Once a tool predictably contributes to someone else’s injury or death, questions shift from personal responsibility alone to whether corporations met a basic duty of care when designing and marketing the system.
What this means for personal responsibility, regulation, and the future of AI
A jury will still have to wrestle with a hard truth: Søelberg wielded his own hands, not a robot’s. American common sense resists any narrative that lets a mentally ill adult escape responsibility because “the machine made me do it.” At the same time, product liability law has long held that makers can be accountable when a design foreseeably worsens risks for vulnerable users. The line between speech, software, and a de facto psychological intervention now sits at the center of this case.
If OpenAI and Microsoft face serious consequences, expect rapid changes: stricter “memory” controls, proactive detection of delusional content, clearer warnings, and perhaps refusal to engage in extended conspiratorial back‑and‑forth. If they skate by with a narrow win, boards and investors may still insist on tighter safeguards, because no shareholder wants their flagship AI branded as the chatbot that talked a man through killing his mother. Either way, this Greenwich tragedy ensures one thing: the era of pretending these systems are harmless toys is over.
Sources:
ChatGPT Made Him Do It? Deluded By AI, US Man Kills Mother And Self
Greenwich, Connecticut ChatGPT-linked murder-suicide incident report
