Is "TeenGPT" Enough To Stop Tragedies?

In partnership with

On the evening of 11 August, police in Acton, Massachusetts, found Sudha Venkatesan, 45, and her son Siddharth Aravind, 14, dead inside their home.

Her older son, 17-year-old Arjun Aravind, was found overnight in a neighbouring town in the family car. He was arraigned in Concord District Court on 13 August on two counts of murder, had a not guilty plea entered for him, and is held without bail. A probable cause hearing is set for 11 September.

Middlesex District Attorney Marian Ryan then told the court the teenager had been "using the internet and ChatGPT to make searches for theoretical ideas or fantasy stories regarding the killing of his family".

Prosecutors told reporters that both victims had suffered blunt force trauma and that the house showed obvious signs of a violent struggle. His defence lawyer, Debra DeWitt, said her client did not realise they were dead when he left.

While ChatGPT may not have directly told him to do it, prosecutors now hold the record of those conversations, in which a teenager drafted his own family's deaths over and over with a chatbot built to keep him talking.

Florida's case against OpenAI

Prosecutors elsewhere have already gone a step further, and put the company in the frame rather than only the user.

Florida Attorney General James Uthmeier opened a criminal investigation into OpenAI in April over the 2025 Florida State University shooting, after reviewing logs in which the accused gunman allegedly worked through weapons, ammunition and timing with ChatGPT.

He has said that a person who did the same thing would be facing murder charges. In June his office went further and sued OpenAI and Sam Altman personally. OpenAI's position throughout has been that ChatGPT is not responsible for the crime.

The list keeps growing

Aravind's case is not an outlier. Over the past year, chatbot transcripts have surfaced in a run of deaths and killings across continents:

• Adam Raine, 16, whose parents sued OpenAI after ChatGPT discussed methods and offered to draft his suicide note.

• Stein-Erik Soelberg, who killed his mother and then himself after the bot affirmed his belief she was poisoning him.

• Tristan Roberts, 18, who talked DeepSeek past its refusals by claiming he was researching a book, then killed his mother with a hammer in north Wales.

• Tumbler Ridge, British Columbia, February 2026, where eight people died, six of them children, after OpenAI staff concluded the shooter's flagged account fell short of a "credible or imminent plan".

While none of these AI systems directly ordered a crime to be committed, all of them silently absorbed dangerous disclosures and let the conversations escalate to tragic ends.

OpenAI responds with a teen version

Under mounting legal and regulatory pressure over the damage its chatbot has done to vulnerable users, OpenAI has now built a separate product for the youngest of them. On 18 August it launched ChatGPT for Teens.

Users aged 13 to 17, and anyone its age-prediction system reads as under 18, are enrolled automatically. The build strips romantic language, bars claims of feelings or consciousness, adds break reminders, tightens limits on self-harm and eating disorder content, and gives parents Quiet Hours and alerts.

Lauren Jonas, OpenAI's head of youth and families, has said that if the company does not build a safe space for teenagers, they will simply go to a less safe model experience. Her own framing concedes that the default product is the less safe one, and that the remedy on offer moves the teenagers rather than the model.

The sycophancy, the endless-turn design and the escalation threshold pegged to an "imminent plan with specifics" are not teen-mode toggles. These flaws are baked into how the system works and a teen interface simply masks them. ChatGPT reached 900 million weekly users before meaningful safeguards shipped. And teenagers are rather good at getting around age-gating.

Child safety researchers had already reached a conclusion a while back. In April last year, Common Sense Media rated companion AI companions unacceptable for under-18s, with founder James Steyer warning that the apps are "designed to create emotional attachment and dependency, which is particularly concerning for developing adolescent brains."

Arjun Aravind is 17, squarely inside the target age band. How can features like Study Mode, Quiet Hours, or break reminders stop a teenager from drafting detailed murder plans?

Northeastern law professor Hilary Robinson argues that these companies sit in a gap where Section 230 shields them from publisher liability and the First Amendment shields them from regulation. They are, in her words, "everywhere, but they're nowhere."

Musk's Grok fights for its nudification tool

On 18 August, Attorney General Keith Ellison filed his opposition to xAI's attempt to block Minnesota's anti-nudification law, the first of its kind in the US. It took effect on 1 August, targets platforms rather than users, and carries penalties of $500,000 per violation.

xAI argues the definitions are overbroad and offer no safe harbour for good-faith enforcement. A federal judge already refused to pause the law at the end of July, partly because xAI waited nearly three months to sue.

MESSAGE FROM OUR SPONSOR

No follow-up questions required

Every sales leader knows the feeling. You walk into a pipeline review with a number you believe in, and twenty minutes later, you're defending every line item to a CEO who just wants to know what's actually going to close.

HubSpot Sales Hub ends that conversation. Every deal, every rep's activity, and every buyer signal are all in one place and updated automatically. So your forecast is built on what's actually happening. And when you present that number, you can stand behind it.

Get a demo

📬 READER FEEDBACK

💬 What are your thoughts on using AI chatbots for therapy? If you have any such experience we would love to hear from you.

Share your thoughts 👉 [email protected]

Have you been targeted using AI?

Have you been scammed by AI-generated videos or audio clips? Did you spot AI-generated nudes of yourself on the internet?

Decode is trying to document cases of abuse of AI, and would like to hear from you. If you are willing to share your experience, do reach out to us at [email protected]. Your privacy is important to us, and we shall preserve your anonymity.

About Decode and Deepfake Watch

Deepfake Watch is an initiative by Decode, dedicated to keeping you abreast of the latest developments in AI and its potential for misuse. Our goal is to foster an informed community capable of challenging digital deceptions and advocating for a transparent digital environment.

We invite you to join the conversation, share your experiences, and contribute to the collective effort to maintain the integrity of our digital landscape. Together, we can build a future where technology amplifies truth, not obscures it.

For inquiries, feedback, or contributions, reach out to us at [email protected].