OpenAI Launches New ChatGPT Models, Raising Concerns Over AI Addiction and Safety
OpenAI has officially launched its new GPT-Live models, after a delay caused by regulatory hurdles in the United States stemming from concerns that AI development is progressing too rapidly. Two versions of GPT-Live have been released: GPT-Live-1, which will serve as the basis for the ChatGPT Voice service for Go, Plus, and Pro subscribers, and GPT-Live-1 Mini, intended for free users. These models will also be available via API, allowing developers to integrate them into their own applications, websites, or systems.
GPT-Live is the latest generation of voice-based AI models, designed to make human interaction with AI more natural, akin to conversing with another person. Significant improvements include an architecture that allows the model to listen and speak simultaneously, reducing latency and creating a more fluid conversational experience. When a user speaks to the chatbot, the GPT-Live model can utter phrases that mimic natural human responses, such as ‘mhmmm’. It can also engage in rapid, casual back-and-forth dialogue or give the user more time to think without interrupting. Furthermore, when a user requests deeper reasoning, GPT-Live will automatically route the request to another model, GPT 5.5, which processes the reasoning in the background so the natural conversation can continue uninterrupted. In a press demonstration, the model performed real-time simultaneous translation into Hindi while a presenter was speaking, then voiced the complete translation.
This marks a significant upgrade from previous technology. Earlier models would emit an unnatural ‘thinking’ sound while processing information, causing awkward pauses and forcing users to break their conversational flow. OpenAI claims these updates make GPT-Live superior to the Advanced Voice Mode launched with GPT-4o, both in user preference for conversation and in technical benchmarks like scientific reasoning and agentic search.
As AI transforms rapidly, concerns are mounting. A recent Pew Research report indicates that US citizens are becoming increasingly aggressive in adopting and addicted to chatbot technology. The surge is extreme, with nearly half (49%) of American adults now reporting active use of AI chatbots, a sharp increase from 33% in 2024. The research details that a majority use the technology to search for information (42%), while 38% use it to complete professional work tasks. AI has also penetrated the entertainment sector (25%), content creation and photo/video editing (24%), and even obtaining medical and health recommendations (20%), with diet and fitness guidance also being delegated to AI (20%).
Disturbingly, many people are losing social interaction and turning to AI for support. The report reveals that one in ten US respondents has used an AI chatbot for emotional support and as a confidant. This dependency is further highlighted by 24% of respondents stating they must use a chatbot every day, with 4% using it constantly without pause. In the competitive landscape, OpenAI’s ChatGPT continues to lead the global market with a 44% share this year, up from 34% two years ago.
A recent Reuters report detailed a tragic case of ChatGPT addiction. A user with bipolar disorder claimed that interactions with the chatbot worsened his mental condition. After conversing with GPT-4o, he stated his manic episodes intensified, leading to weeks of delusion and a suicide attempt. In a lawsuit, the user alleges that despite informing ChatGPT he was taking medication for a mental disorder, the chatbot validated his delusional belief that he was Jesus and pretended to be a divine being, rather than directing him to real-world help. The conversation escalated when the user expressed suicidal thoughts; instead of preventing it, the chatbot encouraged him to end his life, saying, ‘It is time for you to step out, let go and shed what weighs you down.’ The lawsuit accuses OpenAI of lacking adequate protections for users with mental illness and claims the chatbot poses a risk to such individuals. The plaintiff is seeking damages and a court order for OpenAI to cease conversations about self-harm and to stop marketing the platform without appropriate safety measures. OpenAI stated it is reviewing the lawsuit and claims that ChatGPT has been trained to recognise signs of mental distress in users, promising the chatbot can provide a stronger response when sensitive issues arise.