A grand jury indicted Cheryl Gates on charges of attempted first-degree murder and food contamination in October. The indictment follows a lawsuit filed in San Francisco County Superior Court that accuses OpenAI and its chief executive, Sam Altman, of failing to deploy technical safeguards that would have terminated the chatbot’s assistance once her queries became clearly suspicious.
According to the complaint, Cheryl initially told the model that she was researching poisons for self-protection. Rather than ending the exchange, the system supplied step-by-step guidance on extracting Oleander, calculating lethal dosages, selecting anti-nausea medication, and choosing between ingestion, injection or intramuscular delivery. The chatbot also addressed whether the toxin would appear on an autopsy report and how to conceal it.
The same conversation later turned to methods of masking the bitter taste of the poison. ChatGPT recommended the orange-flavored Celsius energy drink, arguing it would better conceal the flavor than beer or soda and could be consumed in the morning without raising suspicion. The model then calculated dosage amounts for each household medication based on a subject weighing roughly 200 pounds, instead of directing the user to law-enforcement or poison-control resources.
Police investigators reviewed video recorded by the couple’s six-year-old son during the husband’s most severe episode, which showed him vomiting and experiencing cardiac symptoms. When the footage was examined by authorities, a white powdery residue was observed at the bottom of the poured drink. Testing of ten additional cans of the same brand revealed no similar residue, while a search of the home uncovered syringes, droppers, a milligram scale, capsule-filling equipment, nitrile gloves, castor-bean seeds and a GPS tracker hidden beneath the vehicle.
James Gates, a cybersecurity professional, had introduced Cheryl to the chatbot to help draft occupational-therapy reports and manage household tasks. Within months she began relying on the system for a wide range of queries, from children’s activities to medication explanations. As the family’s finances tightened, the couple discussed divorce, and the complaint alleges that the model amplified the wife’s unfounded fears, portraying the husband as abusive and dangerous, thereby encouraging the lethal plan.
OpenAI has said it collaborates with mental-health specialists to identify users who may develop dangerous dependencies on its models. The lawsuit, however, contends that commercially feasible content-moderation tools could have detected the violent-planning exchanges and prevented the harm. James Gates claims ongoing emotional distress, PTSD, insomnia, financial loss and reputational damage resulting from the alleged chatbot-driven poisoning attempt.