Developers beware: Google’s Gemma model controversy exposes model lifecycle risks
The recent controversy surrounding Googling‘s Gemma model has once again highlighted the dangers of using test models for developers and the fleeting nature of model availability.
Google removed it Gemma 3 model from AI Studio following a statement from Senator Marsha Blackburn (R-Tenn.) that the Gemma model deliberately hallucinated falsehoods about her. Blackburn said the model made up news stories about her that go beyond “harmless hallucinations” and act as a defamatory act.
In response, Google posted on X on October 31 that it will remove Gemma from AI Studio, saying this is “to avoid confusion.” Gemma remains available via API.
It is also available through AI Studio, which, as the company describes, “is a developer tool (to use it you essentially have to prove you are a developer). We have now seen reports of non-developers trying to use Gemma in AI Studio and asking factual questions. We never intended for this to be a consumer tool or model, or to be used in this way. To avoid this confusion, access to Gemma is no longer available on AI Studio.”
To be clear, Google has the right to remove its model from its platform, especially if people have discovered hallucinations and falsehoods that could spread. It also underlines the danger of relying primarily on experimental models and why enterprise developers should save projects before retiring or deleting AI models. Technology companies like Google still face political controversies, which often affect their implementation.
VentureBeat reached out to Google for additional information and referred them to their October 31 posts. We also reached out to Senator Blackburn’s office, who reiterated her position in a statement that AI companies “must close down.” [models] until you have control over it.”
Experiments with developers
The Gemma family of models, including one 270M parameter versionis best suited for small, fast apps and tasks that can be run on devices like smartphones and laptops. Google said the Gemma models are “built specifically for the developer and research community. They are not intended for actual support or consumer use.”
Nevertheless, non-developers can still access Gemma as it is located on the AI Studio platforma more beginner-friendly space for developers to play with Google AI models compared to Vertex AI. So even if Google never intended Gemma and AI Studio to be accessible to, say, congressional staffers, these situations could still arise.
It also shows that as models continue to improve, these models still produce inaccurate and potentially harmful information. Companies must continually weigh the benefits of using models like Gemma against their potential inaccuracies.
Continuity of projects
Another concern is the control AI companies have over their models. The adage “you own nothing on the internet” remains true. If you don’t have a physical or local copy of the software, you can easily lose access to it if the company that owns the software decides to take it away. Google did not clarify with VentureBeat whether current projects on AI Studio powered by Gemma will be saved.
In the same way, OpenAI users were disappointed when the company announced this would happen remove popular older models on ChatGPT. Even after walking back his statement and Restore GPT-4o back to ChatGPT, Sam Altman, CEO of OpenAI, continues to raise questions about maintaining and supporting the model.
AI companies can and should remove their models if they produce harmful results. AI models, no matter how mature, remain a work in progress, constantly evolving and improving. But because they are experimental in nature, models can easily become tools that tech companies and lawmakers can use as leverage. Enterprise developers should ensure their work can be saved before removing models from platforms.
