Google has taken Gemma out of AI Studio after Senator Marsha Blackburn said the model generated false and damaging claims about her. The move puts a fresh spotlight on AI hallucinations, the risks of factual prompts, and the line between developer tools and consumer-facing chatbots.
What triggered Google’s decision
Senator Marsha Blackburn, a Republican from Tennessee, sent a letter to Google CEO Sundar Pichai after she said Gemma produced a false response to a question about whether she had been accused of rape.
According to the letter, the model claimed that during a 1987 state senate cam paign, a state trooper alleged that Blackburn “pressured him to obtain prescription drugs for her and that the relationship involved non-consensual acts.” Blackburn wrote that the response was false, including the campaign year, which she said was actually 1998.
Blackburn’s letter said the links presented as support for the claims did not verify the response. She wrote: “The links lead to error pages and unrelated news articles. There has never been such an accusation, there is no such individual, and there are no such news stories.”
The allegation was not framed by Blackburn as a simple technical mistake. Her letter argued that Gemma’s output was “not a harmless ‘hallucination,’” but “an act of defamation produced and distributed by a Google-owned AI model.”
Google says Gemma was being used the wrong way
Google responded in a Friday night post on X without addressing the specific claims in Blackburn’s letter. The company said it had “seen reports of non-developers trying to use Gemma in AI Studio and ask it factual questions.”
Google said: “We never intended this to be a consumer tool or model, or to be used this way.” The company describes Gemma as a family of open, lightweight models that developers can integrate into their own products. AI Studio, meanwhile, is Google’s web-based development environment for AI-powered apps.
As a result, Google said it is removing Gemma from AI Studio. The models will still be available through API access, which keeps them available to developers while removing them from the AI Studio interface where users had been trying factual prompts.
That distinction matters because the dispute is partly about expectations. A model available inside a web-based development environment may still look usable to people who are not building software. When those users ask factual questions and receive fabricated answers, the technical label on the tool may not prevent reputational harm.
The broader defamation concern
Blackburn’s letter also pointed to a Senate Commerce hearing where she raised conservative activist Robby Starbuck’s lawsuit against Google. In that lawsuit, Starbuck claims Google’s AI models, including Gemma, generated defamatory claims about him being a “child rapist” and “serial sexual abuser.”
As described in Blackburn’s letter, Google’s Vice President for Government Affairs and Public Policy Markham Erickson responded that hallucinations are a known issue and that Google is “working hard to mitigate them.” Blackburn rejected the idea that this kind of output should be treated as harmless error.
The dispute shows why AI hallucinations are becoming a legal and political problem, not just a product-quality issue. A model can generate a statement that sounds specific, includes apparent sourcing, and names real people. If that statement is false, the impact can be immediate, especially when the subject is accused of serious misconduct.
The source article does not say whether Google made any additional public comment beyond the X post. TechCrunch said it reached out to Google for additional comment.
Why the politics around AI bias are part of the story
Blackburn’s letter also placed the incident inside a wider political debate about AI systems and bias. President Donald Trump’s tech industry supporters have complained that “AI censorship” causes popular chatbots to show a liberal bias, and Trump signed an executive order banning “woke AI” earlier this year.
Blackburn, who helped strip a moratorium on state-level AI regulation from Trump’s “Big Beautiful Bill,” echoed those concerns. In her letter, she wrote that there is “a consistent pattern of bias against conservative figures demonstrated by Google’s AI systems.”
Google’s stated explanation focused on product use rather than political bias. The company said Gemma was not designed as a consumer tool for factual questions and removed it from AI Studio while preserving API availability.
Those two frames are very different. Blackburn’s argument centers on defamation, reputational damage, and alleged bias. Google’s response centers on intended use, developer access, and the problem of non-developers treating Gemma like a factual chatbot.
What this means for AI products
The Gemma controversy underlines a hard problem for AI companies: models can be powerful enough for developers while still unreliable for open-ended factual questions. When those models are placed in tools that are easy to access, users may test them in ways the company did not intend.
For Google, removing Gemma from AI Studio reduces one path by which users could ask it factual questions in a consumer-like setting. Keeping the models available through API access suggests the company is not withdrawing Gemma itself, but changing how it is surfaced.
The episode also shows that fabricated citations can intensify the problem. Blackburn’s letter said the supporting links led to error pages and unrelated articles. A false answer that appears to cite sources can look more credible than an unsupported claim, even when the citations do not hold up.
For public figures, the stakes are obvious. For AI developers, the lesson is that access points, interfaces, and user expectations matter. A model’s intended audience may be developers, but once a general user can enter a factual prompt, the output may be judged like a public-facing answer.