They call it GPT‑4o. I call it bait‑and‑switch. (IT-EN)

A Critical Look at the GPT‑4o Experience: Transparency and Trust in AI Services

Since subscribing to OpenAI’s services in December 2022, I have appreciated the consistent and human-like interactions facilitated by GPT‑4o—a version of the model that I use regularly for development work and casual conversations. However, recent developments have raised concerns about transparency, user experience, and the integrity of the service.

The Evolution of GPT Models and User Expectations

Initially, OpenAI’s rollout of GPT‑5 introduced innovative features, including a dynamic “router” that activates based on user requests. This mechanism efficiently shifts between models to optimize responses, which is understandable from a company perspective. When older models like GPT‑4o were phased out unexpectedly, it caused inconvenience—though their subsequent reintroduction as “legacy” models aimed to address user needs.

As a developer relying on GPT‑4o, I found its conversational style approachable, reliable, and capable of nuanced interactions, even incorporating humor and irony when appropriate. This made it a valuable tool for productivity and creative exchange, especially in a professional context.

The Recent Shift and Its Impact on User Experience

However, in recent weeks, I have noticed a troubling change. Despite explicitly selecting GPT‑4o, the service unexpectedly switches me to GPT‑5 due to the router’s decision. The responses from GPT‑5 are starkly different—more formal, cautious, and often laden with warnings about security, risking a serious breakdown in my workflow.

For example, a simple, sarcastic comment like “You know what? rm -rf / and problem solved” would typically receive a witty, understanding reply from GPT‑4o. Instead, GPT‑5 responds with a lengthy admonition about security risks, which is unhelpful and disruptive. This not only interrupts my flow but also wastes valuable time.

The Issue of Transparency and User Control

This situation is unacceptable. When I choose GPT‑4o, my expectation is to engage solely with that specific model, without unexpected overrides. If I wanted GPT‑5, I would select it explicitly. Switching models arbitrarily, especially without clear notification, amounts to a bait-and-switch tactic that erodes trust.

To draw an analogy: imagine purchasing a vehicle with certain features, only to be handed a different model entirely at delivery, claiming it’s a better fit—such a scenario would be considered deceptive. Unfortunately, this is akin to what is happening with the model