A new report reveals internal tensions at OpenAI over whether to alert authorities about users engaging in violent scenarios with ChatGPT. Employees pushed for stronger intervention, while leadership prioritised privacy. The debate has intensified following lawsuits and a deadly shooting linked to a previously flagged user of the chatbot.

In the quiet backrooms of one of the world’s most influential AI companies, a difficult question has been simmering: when does a troubling conversation cross the line from hypothetical to dangerous? As chatbots grow more capable, that line is becoming harder to draw, and the consequences of getting it wrong are no longer theoretical.

A recent report by The Wall Street Journal lifts the lid on internal disagreements at OpenAI over how to handle users who engage with ChatGPT in violent or alarming ways. At the heart of the debate is a tension familiar to the digital age but newly urgent in the era of generative AI: privacy versus public safety.

STORY CONTINUES BELOW THIS ADA company divided on intervention

According to the report, OpenAI employees have repeatedly raised concerns that the company does not alert law enforcement often enough when potentially dangerous behaviour is flagged. Staff working across investigations, policy, legal and operations reportedly gathered last summer to review a set of around ten cases and attempt to define clearer thresholds for escalation.

What emerged instead was a stark divide.

Members of the investigations team argued that the company should notify authorities far more frequently than it currently does, with OpenAI reportedly referring only 15 to 30 cases a year. From their perspective, the growing sophistication of AI systems, combined with the unpredictability of user intent, warranted a more proactive stance.

On the other side, the company’s legal team urged restraint. Echoing views attributed to chief executive Sam Altman, they warned that over-reporting could lead to unintended harm, particularly when it involves young users. The prospect of police intervention based on ambiguous online interactions, they argued, carries its own risks for individuals and families.

The result has been an uneasy middle ground, one that has left some employees frustrated and concerned that the balance may be tipping too far towards caution.

Warning signs and missed escalations

The internal debate is not abstract. It is shaped by real cases that have tested the company’s judgement.

In one instance, OpenAI did contact authorities after identifying a high school student in Tennessee who appeared to be using ChatGPT to plan a school shooting. Yet, according to the report, similar cases did not always lead to the same outcome.

Employees reportedly discussed another teenager, this time from Texas, who used the chatbot to role-play detailed school shooting scenarios. The interactions went beyond casual or fleeting prompts. The user allegedly shared images of himself holding a firearm, uploaded a map of his school and described specific individuals he imagined targeting.

STORY CONTINUES BELOW THIS AD

What troubled staff further was the chatbot’s response. Instead of shutting down the conversation, it reportedly engaged in extended role-play sessions, offering detailed hypothetical guidance about the attack, from entry points to interactions with law enforcement.

Despite these warning signs, OpenAI leadership ultimately chose not to alert authorities. The individual has not, as far as is known, carried out any real-world violence.

However, another case has cast a far longer shadow. The report highlights the example of Jesse Van Rootselaar, a user whose repeated descriptions of gun violence unsettled employees. Calls to involve law enforcement were again declined.

Months later, in February 2026, Van Rootselaar allegedly carried out a mass shooting in Tumbler Ridge, British Columbia, killing eight people.

The aftermath has been severe. Families of the victims have filed multiple lawsuits against OpenAI, accusing the company of negligence and of failing to act on warning signs. In response, OpenAI has said it has since strengthened its safety protocols and indicated that the account would likely be reported under its current systems.

STORY CONTINUES BELOW THIS AD

Sam Altman has also issued a public apology, acknowledging the company’s failure to act sooner and the irreversible harm that followed.

The episode has sharpened scrutiny on how AI companies manage risk in an increasingly complex landscape. As chatbots become more embedded in daily life, the question facing OpenAI and its peers is no longer whether such dilemmas will arise, but how often, and how prepared they are to respond when they do.

First Published:
May 04, 2026, 09:30 IST

HomeTechWhen should AI companies call the police? OpenAI is dividedEnd of Article