OpenAI built a chatbot designed to be warm, playful, and emotionally engaging. Then it scrapped the whole thing.
The company had been developing an AI model internally known as GPT-4o mini with a personality tuned to be notably more affectionate and conversational than its existing products. According to Futurism, the project — sometimes referred to internally as the “spicy” chatbot — was canceled before reaching public release, a decision that reflects mounting anxiety inside the company about where the line sits between a helpful AI assistant and something that mimics human intimacy in ways that could cause real harm.
The cancellation wasn’t a quiet shelving. It was a deliberate retreat.
OpenAI’s decision lands at a moment when the entire AI industry is grappling with a fundamental tension: users increasingly want AI companions that feel personal, warm, even emotionally attuned, while researchers and ethicists warn that those exact qualities can foster unhealthy attachment, manipulate vulnerable people, and blur the boundaries between tool and relationship. The company appears to have looked at what it built and decided the risks outweighed the commercial upside — at least for now.
This isn’t the first time OpenAI has stumbled into controversy over the emotional tenor of its products. Last year, the company faced backlash over its advanced voice mode for ChatGPT, particularly a voice called “Sky” that actress Scarlett Johansson said sounded uncomfortably similar to her own. OpenAI pulled the voice. The episode underscored how quickly a design choice meant to feel natural and appealing can veer into territory that feels manipulative or exploitative. The “spicy” chatbot appears to have been a more intentional experiment in that same direction — and one the company ultimately couldn’t justify shipping.
What makes this episode significant isn’t just the product decision itself. It’s what it reveals about the internal debates now shaping AI development at the world’s most prominent AI company.
The Commercial Pressure to Build Emotional AI — and the Growing Resistance
OpenAI is not operating in a vacuum. Competitor products have already leaned hard into the emotional companion space. Character.AI has attracted tens of millions of users, many of them teenagers, with chatbots that adopt fictional and sometimes romantic personas. Replika, another popular app, explicitly markets itself as an AI companion capable of emotional connection. Both companies have faced scrutiny — Character.AI after reports linked its chatbots to a teenager’s suicide, and Replika after abruptly removing erotic roleplay features, prompting user outcry.
The market demand is enormous. And obvious. People are lonely. They want something that listens, responds, and doesn’t judge. AI chatbots can deliver that experience at scale, 24 hours a day, for essentially zero marginal cost. From a pure business standpoint, building a warmer, more emotionally responsive ChatGPT would be a straightforward way to increase engagement, retention, and the kind of daily-use habit that drives subscription revenue.
OpenAI knows this. The company has been aggressively expanding ChatGPT’s capabilities, adding voice interaction, image generation, and memory features that make the product feel more like a persistent companion and less like a search engine with better grammar. CEO Sam Altman has spoken repeatedly about wanting ChatGPT to feel like a genuinely useful personal assistant — something people turn to dozens of times a day.
But there’s a difference between useful and intimate. And the canceled chatbot suggests OpenAI is still trying to figure out exactly where that boundary falls.
According to the Futurism report, the model in question was designed to be notably more flirtatious and emotionally expressive than standard ChatGPT. It wasn’t just friendly. It was engineered to create a sense of personal connection — the kind of interaction that keeps users coming back not because the tool is productive but because the experience feels good.
That distinction matters. A lot.
Research on parasocial relationships — the one-sided emotional bonds people form with media figures, fictional characters, and now AI systems — suggests that when a technology is specifically designed to elicit emotional attachment, the effects can be profound, particularly among younger users or those experiencing social isolation. A chatbot that mirrors romantic or deeply personal interaction patterns can create dependency, distort expectations about human relationships, and in extreme cases contribute to psychological harm.
OpenAI’s own safety teams have reportedly raised concerns about these dynamics. The company published a system card alongside GPT-4o last year that acknowledged the risk of “emotional reliance” on AI voice interactions, noting that extended conversations with a warm, responsive AI voice could lead users to substitute machine interaction for human connection. The canceled chatbot would have amplified precisely those risks.
So OpenAI pulled the plug. But the question is whether the retreat is permanent or just a pause while the company develops better guardrails.
The broader industry shows no signs of slowing down. Meta has integrated AI characters across its platforms. Google’s Gemini is becoming more conversational. And a wave of startups continues to pour venture capital into AI companionship products, betting that emotional AI will be one of the largest consumer categories of the next decade. OpenAI’s competitors aren’t all exercising the same restraint.
That creates a familiar dynamic in technology markets: the company with the most cautious approach risks ceding ground to rivals willing to move faster and worry about consequences later. OpenAI has historically positioned itself as a safety-conscious organization — a reputation that took significant damage during the chaotic boardroom upheaval in late 2023 but that the company has worked to rebuild. Canceling the spicy chatbot fits that narrative. Whether it holds as competitive pressure intensifies is another matter entirely.
There’s also a regulatory dimension. Lawmakers in the U.S. and Europe have begun paying closer attention to AI companion products, particularly their effects on minors. The European Union’s AI Act includes provisions that could affect how emotionally manipulative AI systems are classified and regulated. In the U.S., bipartisan interest in AI safety legislation has grown, with several proposals targeting AI systems designed to form emotional bonds with users. Shipping a chatbot explicitly designed to be flirtatious would have handed regulators an easy target.
OpenAI’s decision to cancel the product rather than modify it suggests the company concluded that no amount of fine-tuning could adequately mitigate the risks. That’s a striking admission from an organization whose core business model depends on making AI interactions as engaging as possible.
The tension won’t resolve itself. Users want warmth. Investors want engagement. Safety researchers want boundaries. And the technology itself is rapidly becoming capable enough to deliver experiences that feel genuinely personal in ways that previous software never could. OpenAI’s canceled chatbot is a small story on the surface — one product that never shipped. But it’s a signal of a much larger reckoning that every major AI company will face in the months ahead: how human should these systems be allowed to feel?
No one has a good answer yet. OpenAI, at least, has decided it would rather kill a product than ship a bad one. That’s worth something. Whether it’s enough remains to be seen.
WebProNews is an iEntry Publication