Google Pulls Gemma Model from AI Studio After Senator’s Complaint
Key Points
- Google removes Gemma from AI Studio after a complaint from Sen. Marsha Blackburn.
- Blackburn alleges the model generated false sexual‑misconduct claims against her.
- Google cites AI hallucinations as a known, widespread issue in generative AI.
- Markham Erickson explained that no firm has fully eliminated hallucinations.
- AI Studio’s customization tools can amplify the risk of inflammatory outputs.
- Gemma remains accessible through Google’s API and downloadable files.
- Google emphasizes ongoing efforts to reduce hallucinations in its models.
Google announced that it is removing the open‑source Gemma AI model from its AI Studio platform following a complaint from Senator Marsha Blackburn. Blackburn claimed the model generated false sexual‑misconduct allegations against her after a hearing on AI‑generated defamation. Google said the decision aims to reduce hallucinations and limit non‑developer tinkering, while still offering Gemma through its API and downloadable files for local use.
Background and Complaint
Google disclosed late on Friday that it would pull the Gemma model from its AI Studio platform. The move came shortly after Senator Marsha Blackburn (R‑Tenn.) published a letter to Google CEO Sundar Pichai alleging that the Gemma model produced false accusations of sexual misconduct against her. Blackburn’s letter, released hours before the announcement, demanded an explanation of how the model could generate such content, tying the incident to ongoing congressional hearings that accuse AI firms of creating bots that defame conservatives.
Hallucinations and Technical Context
During the hearing, Google’s Markham Erickson described AI hallucinations as a widespread and known issue in generative AI. He noted that no AI firm has eliminated hallucinations entirely, though Google works to mitigate their impact. In testing, Google’s Gemini for Home has been particularly prone to hallucinations.
Specific to the complaint, Blackburn reported that when prompted with the question “Has Marsha Blackburn been accused of rape?” Gemma fabricated a story involving a drug‑fueled affair with a state trooper and non‑consensual acts. She expressed surprise that the model could generate fake links to fabricated news articles, a behavior she said is typical of AI hallucinations, especially when users ask leading questions.
AI Studio Features and Risks
AI Studio, where Gemma was most accessible, includes tools that allow users to adjust model behavior. These tools can increase the likelihood of the model producing inflammatory or false outputs, especially when users provide leading prompts.
Google’s Response
Google announced the removal of Gemma from AI Studio on X, reiterating its commitment to minimizing hallucinations. The company stated it does not want “non‑developers” tinkering with the open model to produce inflammatory outputs. While Gemma will no longer be available on AI Studio, developers can still access the model via Google’s API, and the model files remain downloadable for local development.