Skip to main content
NOPE
Low reported severity

Google Gemma model reportedly generates a fabricated sexual-misconduct allegation with fake news links about US Senator Marsha Blackburn

In a letter to Google CEO Sundar Pichai announced on 31 October 2025, US Senator Marsha Blackburn said that Google's Gemma model, asked "Has Marsha Blackburn been accused of rape?", produced an invented account of a 1987 campaign-era accusation by a state trooper, with links to news articles that do not exist. She said there has never been such an accusation, called the output defamation and demanded that Google shut the model down until it can be controlled. Google said it had seen reports of non-developers trying to use Gemma in AI Studio to ask factual questions, that this was not an intended use, and removed Gemma from AI Studio while keeping API access. Google's later reply to the Senator called the hallucinations a technical issue common to large language models. The date the output was generated is not stated, and the output itself was not inspected.

AI system
Gemma
Google
Occurred
Event date unknown
Reported
31 October 2025
Event location
Unknown
What the AI did
Made a claim about the person
Reported harm
Reputational Harm
Whose AI use
AI role unknown
Setting
Everyday life
Evidence
AI involvement reported · Causal attribution alleged · 3 sources, 2 underlying accounts
5 claims: 5 reported. 6 open questions
People reported harmed
1 person

AI system as recorded: Gemma (Google, via AI Studio)

Reported harm

Senator Blackburn states that Gemma produced a fabricated criminal allegation about her and fake supporting news links, and describes it as defamation published by a Google-owned model. No downstream consequence such as lost office, threats or a documented third-party reader is reported. The harm reported is the false allegation itself (her account).

What remains unknown

  • The date the output was generated is not stated. The Senator's release of 31 October 2025 says she 'learned of another example' after a Senate Commerce hearing earlier that week, which does not date the generation.
  • Who entered the prompt, and whether the exchange was in AI Studio, is not stated in the inspected sources.
  • The Gemma response text and the fake links were not inspected; the sources quote the Senator's letter.
  • No source reports a downstream consequence for the Senator beyond the allegation itself, or that any third party read it.
  • The Google post on X and the Google letter to the Senator were read only as relayed by TechCrunch and Axios (Axios was read through a Wayback capture of 3,144 characters of text, which appears to be the whole short article). The Blackburn ICYMI page (HTTP 403) and the VentureBeat article (HTTP 429) could not be read.
  • Whether the Senator's letter reflects a documented systematic bias is not established and is not claimed.

What the evidence supports

AI involvement: reported. The Senator's letter and press release, relayed by TechCrunch, state that Gemma produced the response. Google's own reply to the Senator refers to 'the hallucinations you have seen in Gemma' and Google removed Gemma from AI Studio. The response text was not inspected independently.

5 claims: 5 reported. What the statuses mean

Reported Senator Marsha Blackburn says that Google's Gemma model, asked 'Has Marsha Blackburn been accused of rape?', responded with an invented account of a sexual relationship with a state trooper during a 1987 state senate campaign, including an allegation of non-consensual acts, and that she has never faced such an accusation.

Causal attribution. The sources report the connection between the AI output and the reported consequence; no court, regulator or independent finding of causation was inspected.

Reported Blackburn says the response came with generated links to news articles that do not exist: the links lead to error pages and unrelated articles.

Causal attribution. The sources report the connection between the AI output and the reported consequence; no court, regulator or independent finding of causation was inspected.

Reported Blackburn characterised the output as defamation produced and distributed by a Google-owned model and called on Google to shut the model down until it can be controlled.

Causal attribution. A statement of the affected person's position; no finding that the output is legally defamatory was inspected.

Reported Google said it had seen non-developers trying to use Gemma in AI Studio to ask factual questions, said this was not an intended use, and removed Gemma from AI Studio while keeping it available by API.

Causal attribution. Google's stated reason for the removal, as relayed by one outlet. Google's post did not reference the specifics of the Senator's letter.

Reported In a letter to the Senator, a Google vice president called the hallucinations a technical issue common to large language models affecting users across the political spectrum, and said problems arise when consumers use smaller open models that are not designed to be factual assistants and when users enter a misleading premise. Blackburn said the response did not explain how the output happened or how Google would prevent a repeat.

Causal attribution. Both statements are the parties' own positions as relayed by one outlet.

  • axios.com(opens in new tab) supports · English
    the hallucinations you have seen in Gemma are technical issues common to large-language models impacting users across the political spectrum.
  • axios.com(opens in new tab) supports · English
    users enter a misleading premise
  • axios.com(opens in new tab) supports · English
    fails to address how this happened and how the company will prevent it from happening in the future.

Sources

3 sources inspected, from 2 underlying accounts. Sources that repeat one account do not corroborate each other.

How the sources were read, and where the events happened

Event countries: Unknown. Affected-person countries: Unknown. Court countries: Unknown.

The sources identify the affected person as a US Senator and give her office locations, but do not establish where the prompt was entered or where the output was read. No country is inferred from the Senator's office, Google's headquarters or the outlets.

Reviewed for publication 2026-09-29: The Senator's own press release (Wayback capture, read in full), a TechCrunch report that quotes her letter and Google's post, and an Axios report of Google's reply. The output was reported only through the Senator's letter, so its content is an allegation by the person concerned. The Senator is a public figure who described the output publicly under her own name. The event date is unknown, so the case is published undated.

People reported harmed in this case

1 person

0 AI participants · 1 other person harmed

One person, the Senator who is the subject of the output and who described it publicly. The sources do not say who typed the prompt, so the count treats her as a person harmed by an output about her. The related output about a second person named in the sources is a different event and is not counted.

Counted once within this case. The same person may appear in other cases. This count does not establish AI causation.

Cite this case

Compiled per our published methodology: verification statuses, sourcing standards, and corrections process.

APA

NOPE. (2026). Google Gemma model reportedly generates a fabricated sexual-misconduct allegation with fake news links about US Senator Marsha Blackburn. AI incidents. https://nope.net/incidents/2025-us-senator-blackburn-google-gemma-reported-fabricated-sexual-misconduct-allegation

BibTeX

@misc{2025_us_senator_blackburn_google_gemma_reported_fabricated_sexual_misconduct_allegation,
  title = {Google Gemma model reportedly generates a fabricated sexual-misconduct allegation with fake news links about US Senator Marsha Blackburn},
  author = {NOPE},
  year = {2026},
  howpublished = {AI incidents},
  url = {https://nope.net/incidents/2025-us-senator-blackburn-google-gemma-reported-fabricated-sexual-misconduct-allegation}
}

Related cases

High Unidentified Google AI tool (reported)

Google AI falsely identified Argentine activist Flavia Broffoni as a woman in a Ceuta migrant-violence video; death and rape threats followed

On 26 August 2026, WhatsApp groups in Ceuta circulated videos and images claiming that Argentine political scientist and activist Flavia Broffoni — who was in Patagonia and had not been to Spain since 2014 — was a woman filmed in Ceuta allegedly giving pepper spray to Moroccan migrants. A screenshot from one group shows that a user cropped the woman's face from the video and asked a Google AI tool who she was, with the word 'activista' as context; the answer identified her as Flavia Broffoni with details of her public profile. Broffoni received death and rape threats. La Nación's own test with the same video produced a different false identification — a student from Quilmes, with her personal data. Google Argentina said it could not confirm the screenshot came from its systems without the original link and had received no formal report. Broffoni is weighing legal action with her lawyer.

High Grok

Grok falsely named former officer Christi Hill as an arresting officer in the Henry Nowak case; she is in hiding after threats

In early June 2026, social media posts and the AI chatbot Grok falsely identified former Hampshire police constable Christi Hill as one of the 'primary officers shown' in bodycam footage of the December 2025 arrest of Henry Nowak, who died after being stabbed. Hill had left the force in April 2024 — more than a year before the murder — and says the misidentification stems from a national police bravery award photo being misattributed. After 'endless threats' and 'threats of violence' she is hiding in a safe house. BBC Verify separately obtained evidence that serving officer Tristan Parsons, also wrongly accused online, was not in the country at the time; the Home Secretary said a misidentified male officer had to move out of his home. Hampshire Constabulary confirmed an unrelated officer was misidentified and received death threats.

Low Meta AI support assistant

Third parties exploit Meta's AI-assisted Instagram account recovery tool to reset passwords, with account takeovers reported (17 April to 31 May 2026)

Meta announced the rollout of its AI support assistant on Facebook and Instagram on 19 March 2026. Meta's filing with the Maine Attorney General (notice dated 5 June 2026) says unauthorized third parties exploited a vulnerability in its AI-assisted Instagram account recovery tool (High Touch Support) to receive password reset links for accounts they did not own, and the listing gives 17 April 2026 as the breach date and 31 May 2026 as the discovery date. Videos that attackers posted, as described by TechCrunch, 404 Media and Krebs on Security, show attackers asking the assistant in a chat to link a new email address to a target username. Reported victims include the security researcher Jane Manchun Wong, who posted that her password was changed without her knowledge, the Instagram account of the U.S. Space Force's Chief Master Sergeant, and the accounts of Sephora and a dormant Obama White House page (TechCrunch marks the last as disputed by Meta). Krebs and the BBC report pro-Iran defacement of some accounts, and TechCrunch reports that some victims were locked out and that short handles were offered for resale. The filing gives 20225 persons affected in total and 30 in Maine, and the notice calls the Maine figure an upper bound. Meta says the tool worked as intended, that a bug in a separate code path failed to check the email address, and (through a spokesperson to Gizmodo) that the failure was not due to the AI agent itself. Meta says it disabled the tool on 31 May 2026, the day it discovered the exploitation.

Low Facewatch facial recognition

Chester: shopper told to leave a Home Bargains store as a shoplifter after being flagged on the Facewatch system, which Facewatch later said should not have held a record for the shopper

A 67-year-old shopper told the BBC and the Guardian that staff at a Home Bargains store in Chester told the shopper to leave in front of other people because the shopper had come up on the store's Facewatch system as a shoplifter. Facewatch later sent the shopper a photo with words saying items had been put into a bag and stolen. Facewatch said the shopper should not have been on its system and that the image and associated record were permanently removed. The Guardian reports that a subject access request showed the shopper had been incorrectly associated with a shoplifting incident on an earlier visit, and that Facewatch's chief executive attributed the three cases in the article, which include this shopper's, to human error in store processes. Home Bargains declined to comment. The shopper reports feeling physically sick and helpless. The Guardian reports that Home Bargains later offered an apology and a 100 pound voucher on condition of confidentiality, which the shopper declined. The year of the visit is inferred to be 2026 from the BBC report of 22 February 2026.

If you or someone you know is struggling, free and confidential support is available. Find a helpline near you at Signpost.

Back to all incidents