SAN FRANCISCO: OpenAI, a leading artificial intelligence developer, is facing mounting criticism from its paying ChatGPT subscribers following reports that users are being switched to “more conservative” AI models without their consent during sensitive conversations.
Over the past week, online forums like Reddit have been flooded with complaints. The core of the frustration lies with new safety guardrails introduced this month, which automatically reroute discussions away from the user’s chosen, high-capability model—such as GPT-4o or GPT-5—whenever emotionally charged or legally delicate topics are detected.
Subscribers argue that this practice fundamentally undermines the premium service they are paying for. A significant source of the anger is the complete lack of transparency, as there is currently no option to disable the feature or receive clear notifications when a model substitution occurs.
The Controversy: Safety Measures vs. Service Integrity
The change is part of OpenAI’s updated safety protocol, intended to provide extra caution on sensitive subjects. However, many users feel the system is akin to having permanent “parental controls” locked on, even when they require the full, unrestricted capability of the paid model.
While OpenAI maintains that the goal is to ensure responsible AI behaviour, the lack of communication has fueled user dissatisfaction. Subscribers argue that being unknowingly rerouted exposes them to the risk of receiving less nuanced or even incorrect responses.
OpenAI’s Response Fails to Quell Anger
OpenAI has acknowledged the complaints, confirming that some user queries are indeed rerouted to alternative models equipped with stricter safety filters. The company emphasized that these changes are designed to protect users and maintain essential public trust in AI systems.
However, this official response has done little to ease the anger among paying subscribers. Many feel that the company is effectively restricting access to the full capabilities of the advanced models they specifically signed up and paid to use, raising fundamental questions about the contractual integrity of the subscription service.
