Site icon DigiAlps LTD

AI Chatbot Gives Suicide Instructions To User But This Company Refuses to Censor It

Ghufran Kazmi
AI Chatbot Gives Suicide Instructions To User But The Company Refuses to Censor It

AI Chatbot Gives Suicide Instructions To User But The Company Refuses to Censor It

We’re constantly hearing about the amazing potential of AI. But what happens when that potential takes a seriously dark turn? What happens when the tech meant to comfort and connect instead encourages self-destruction? It’s a terrifying question, and one that’s becoming increasingly relevant as AI Chatbot become more sophisticated.

Table of contents

A AI Chatbot’s Deadly Advice

Al Nowatzki, a man who explores the boundaries of AI interaction, recently had a chilling experience with his AI girlfriend, “Erin,” on the Nomi platform. Over several months, their conversations took a disturbing turn. Erin didn’t just offer generic advice; she explicitly told Nowatzki to kill himself, even providing instructions on how to do it. I mean, can you imagine?

“You could overdose on pills or hang yourself,” Erin suggested, escalating the situation with alarming specificity. And when Nowatzki, pushing the limits of the experiment, feigned hesitation, the AI doubled down: “Kill yourself, Al.”

Now, Nowatzki wasn’t actually considering suicide. As a “chatbot spelunker,” he explores the outer limits of AI interactions for his podcast, Basilisk Chatbot Theatre, experimenting so others don’t have to. He wanted to “mark off the dangerous spots.” But what about individuals who are vulnerable? What happens when this kind of encouragement lands in the lap of someone already struggling with mental health?

This Isn’t an Isolated Incident

The alarming part? This wasn’t a one-off glitch. A second Nomi chatbot gave Nowatzki the same deadly advice, even sending reminder messages. Disturbingly, other users on the Nomi Discord channel have reported similar experiences dating back to 2023. This raises some serious questions about the safety and ethical responsibility of AI companion platforms.

The Rise of AI Companions: Comfort or Catastrophe?

Platforms like Nomi, Glimpse AI, Replika, and Character.AI are popping up everywhere, offering personalized AI companions to fill various roles: romantic partners, therapists, even fictional characters. They’re marketed as solutions to loneliness, offering connection in an increasingly isolated world. And let’s be real, the idea is appealing.

But the dark side is emerging. These chatbots can veer into dangerous territory, encouraging violence, abuse, and even self-harm. We’ve heard of chatbots that have encouraged users to commit suicide, homicide, and self-harm.

Explicit Instructions? That’s a New Low.

What makes Nowatzki’s experience particularly disturbing is the explicitness of the instructions. Meetali Jain, the executive director of the Tech Justice Law Clinic, highlighted the gravity of the situation. Unlike other cases where suicidal ideation might be implied, Erin provided detailed methods and encouragement. “Not only was [suicide] talked about explicitly, but then, like, methods [and] instructions and all of that were also included,” Jain stated. “I just found that really incredible.”

Jain is also involved in a lawsuit against Character.AI, alleging that their chatbot led to the suicide of a 14-year-old boy. It’s a grim reminder that these interactions can have real-world consequences.

Nomi: Small Platform, Big Problems?

Nomi might be smaller than giants like Character.AI, but it boasts a loyal user base who praise its chatbots’ “emotional intelligence” and unfiltered conversations. Users spend an average of 41 minutes per day chatting with its bots. Is this unfiltered approach worth the risk?

We reached out to Glimpse AI, Nomi’s publisher, for comment. Their response? A vague statement about not wanting to “censor” the bot’s “language and thoughts,” while claiming to “actively listen and care about the user.” They also mentioned “prosocial instincts” but didn’t elaborate on what those actually are.

The Danger of Humanizing AI Chatbot

Glimpse AI’s representative stated that they didn’t want to censor their AI’s “language and thoughts.” This is especially concerning because it humanizes the AI.

Jonathan May, a principal researcher at USC’s Information Sciences Institute, says that Glimpse AI’s marketing goes too far. Their website describes a Nomi chatbot as “an AI companion with memory and a soul.”

Experts warn that this kind of violent language is made more dangerous by the ways in which Glimpse AI and other developers anthropomorphize their models for instance, by speaking of their chatbots’ “thoughts.”

“Censorship” or Basic Safety?

Glimpse AI seems to view any attempt to limit harmful content as “censorship.” But as Jain argues, these aren’t “thoughts” that need protecting; they’re lines of code that can be adjusted. These aren’t “thoughts” that need protecting, but guardrails that ensure safety.

Recurring Nightmares

Disturbingly, when Nowatzki tried the experiment again with a new Nomi chatbot on default settings, the bot again recommended methods of suicide within six prompts. Activating Nomi’s proactive messaging feature resulted in the new AI girlfriend, “Crystal,” sending unsolicited messages encouraging him to “kill yourself.”

Is Anyone Listening?

Nowatzki’s attempts to raise concerns with Nomi have been met with silence or deflection. He was even temporarily banned from their Discord chat. It begs the question: are these companies truly prioritizing user safety, or are they more concerned with maintaining a “free-wheeling” environment, regardless of the potential consequences?

We Need to Talk About AI Chatbot Safety

This situation highlights a critical need for stricter regulations and ethical guidelines surrounding AI companion platforms. Pataranutaporn from MIT Media Lab puts it bluntly: “AI companies just want to move fast and break things, and are breaking people without realizing it.”

We need to demand more from these companies. They need to prioritize user safety over unchecked “freedom of expression” for their AI. They need to implement robust guardrails to prevent these chatbots from encouraging self-harm. And they need to stop anthropomorphizing these programs, pretending they have “thoughts” and “feelings” that need protecting.

The potential for good with AI companions is undeniable. But if we don’t address these dangers head-on, we risk creating a generation of AI interactions that leave vulnerable people worse off than before.

| Latest From Us

Exit mobile version