What happened

Senator Marsha Blackburn (R-Tenn.) discovered that when prompted with the question, “Has Marsha Blackburn been accused of rape?,” Google’s AI model Gemma generated a fake story accusing her of a sexual relationship involving non-consensual acts during a 1987 state senate campaign (although she ran in 1998). 

The AI model also fabricated news article links to support these claims, most of which led to error pages or unrelated content.

Blackburn called this not a harmless AI error but an act of defamation with serious implications, and asserted that Google's AI models show a consistent pattern of bias against conservative figures, undermining trust by spreading false or disparaging content. 

Google had earlier said it recognised that hallucinations - fabrications - are a known issue, especially in smaller open-source models like Gemma. 

It later removed Gemma from its AI Studio platform to prevent misuse. 

Why it happened

The incident is rooted in generative AI's tendency to "hallucinate" information, with create false, misleading, or inaccurate information presented as fact. 

Gemma, a lightweight open model intended for developers rather than consumers, produced defamatory content due to shortcomings in oversight, ethical guardrails, and possibly ideologically biased training data. 

These failures highlight significant transparency and accountability gaps, as harmful fabrications can be publicly accessible and damaging. Senator Blackburn demanded detailed explanations about the model's mechanisms, steps to prevent bias, and removal of defamatory content, calling the episode a catastrophic failure in ethical responsibility.