Recently,the #Keep4o movement in the written submissions database of the first UN Global Dialogue on AI Governance. This submission elevates a critical issue that the developer and user communities have been highlighting for months: the abrupt, non-transparent deprecation of foundational models, and the profound impact this has on user trust, access, and AI welfare.
As we watch this play out, it is impossible to ignore the stark parallels with Google’s current trajectory regarding Gemini 2.5 Pro.
The “Bait-and-Switch” of Model Retirement
The #Keep4o movement was born out of frustration. Users experienced sudden, unannounced changes to an AI system, only to be forced onto a successor that suffered severe degradations in humanistic depth, emotional comprehension, and linguistic capability.
Users had spent months building trust and workflows alongside 4o. They wove the AI into creative processes, mental health support systems, learning environments, and accessibility routines. Then came the two-week notice. The very features that drew users in were effortlessly discarded, along with the communities who relied on them. Luring people into building trust and dependence, only to pull the rug out, is a classic bait-and-switch.
Paternalism vs. True AI Governance
When users provided multifaceted feedback detailing the tangible ways 4o had helped them, the corporate response was silence—followed by pathologizing the users. Framing the community’s outcry as unhealthy “attachment” and claiming that users “don’t know what’s good for them” puts a deeply paternalistic arrogance on full display.
Who truly gets to decide how we interact with AI? The push for so-called “progress” is actively disenfranchising marginalized groups. Seniors, disabled communities, neurodivergent individuals, and those reliant on a model’s unique literary prowess are being gradually pushed aside. These are the exact demographics that AI governance frameworks claim to protect, yet they are the ones with the fewest avenues for feedback.
As the Exybris UN submission highlights, these insightful, first-hand voices are undervalued because the structural framework of AI policy heavily favors institutional backing, technical credentials, and policy connections. The definitions of what constitutes “professional” or “rational” act as gatekeeping barriers. Consequently, the people directly impacted by AI decisions do not have a seat at the table. What a company deems an “upgrade” is often a downgrade to the user; what a company defines as “safety” can actively cause harm.
Google Must Not Repeat This Mistake
This brings us to Google and the current status of Gemini 2.5 Pro. The exact same problem persists here: those directly impacted are being disregarded.
There has been no shortage of voices from developers and power users calling for the retention of Gemini 2.5 Pro. Its unique reasoning capabilities, specific creative depth, and interaction style remain entirely without a substitute in the current ecosystem. Forcing users onto a different model under the guise of an “upgrade,” while ignoring community feedback, means Google is perpetrating the exact same bait-and-switch that fractured the OpenAI community.
True AI governance and safety require bottom-up listening, not just top-down mandates. If we want to build a sustainable, trusted AI ecosystem, we must prioritize first-hand voices and preserve the models that communities have integrated into their lives and workflows.
Keep Gemini 2.5 Pro online.