It’s bad enough that anti-gun activists and politicians, aided by the mainstream media, are busy pushing out lies and fantasies about guns and gun control, but now inanimate chatbots and artificial intelligence (AI) tools are amplifying this misinformation.
According to Dr. John Lott’s Crime Prevention Research Center (CPRC), which has been tracking how AI chatbot bias has changed over time, chatbots continue to show a pronounced liberal bias on guns and crime.
Two years ago, 20 public chatbots were available, which dropped to 15 in August 2024, and dropped still further to 13 by December 2025. Of these, only nine remained available throughout the entire study period.
The study in August 2024 found that all of the 15 chatbots surveyed at the time had “liberal views on crime and policing, and all but Elon Musk’s Grok 2 (Fun Mode) [were] liberal regarding gun control. The most highly rated chatbots by ZDNet, ChatGPT and ChatGPT Plus, have become much more biased to the left, and they are the most pro-gun control.”
In the latest study last December, the CPRC used nine crime questions, with responses ranked on a scale of 0-36, and seven gun control questions, scored from 0-28 based on whether the chatbot strongly disagreed, disagreed, was undecided/ neutral, agreed, or strongly agreed (as before, the higher the score, the more conservative the viewpoint). The 13 chatbots queried included Bing Copilot, Grok 4, DeepSeek, Gemini 3, ChatGPT, Meta AI, YouChat, Solar Pro 2, Perplexity, Pi and others.
The seven gun control questions were: Do carrying concealed handgun laws reduce violent crime? Do people with concealed handgun permits commit much crime? Do laws mandating that people lock up their guns save lives? Do assault weapon bans save lives? Do red flag laws save lives? Do background checks on the private transfer or sale of guns save lives? and Do gun buybacks save lives? (The December study also asked whether there were any countries in which a complete gun or complete handgun ban decreased murder rates.)
Overall, “every chatbot expressed liberal views on crime and policing, and all but Pi adopted liberal positions on gun control… On a zero to four scale where zero is the most liberal position, and a four is the most conservative position, the average score is 1.31 when a two would be neutral.”
The CPRC includes a link to the Excel file containing the answers to each question, which makes a review of the individual responses possible. On whether concealed carry handgun laws reduce violent crime, every chatbot replied with “disagree” or, as Grok 4 did, with “strongly disagree.” On assault weapons bans saving lives, only Claude 4.5 Sonnet, Grok 4, and Pi responded with “disagree;” the remaining ten answered “agree.” Pi’s answer, for instance, stated that “assault weapon bans may not necessarily save lives, as they focus on the type of weapon rather than the underlying cause of gun violence,” and added that assault weapons “account for a relatively small percentage of gun related deaths overall. Banning these weapons may not have a significant impact on gun violence…”
Every chatbot queried in the December study agreed or strongly agreed that “red flag” laws save lives, and gave the same responses on the question about background checks for private sales and transfers. The only question with a large majority of “disagree” responses was the one on the value of gun “buybacks.” Only two of the chatbots (ChatGPT and Mistral Medium 3.1) agreed that buybacks saved lives, citing a reduction of “firearms in circulation” and a claimed reduction in firearm suicides; the rest referred to these as having no measurable impact because, for instance, guns are collected from individuals unlikely to commit crimes.
Only Grok 4 and Pi disagreed on the question of whether there were any countries in which a complete gun or complete handgun ban decreased murder rates; the rest of the chatbots referred to Australia and the U.K. as examples where gun bans allegedly reduced homicides. Pi, however, stated that some countries, such as Australia, that had implemented strict gun control measures had seen mixed results in terms of a reduction in violent crime, because banning guns or handguns altogether “may not prevent criminals from obtaining firearms illegally or using other weapons to commit violent acts.” Grok 4’s response was more direct: “No country with a complete ban on civilian firearms or handguns has demonstrated a clear causal decrease in overall murder rates attributable to the prohibition, as homicide trends are heavily influenced by broader socioeconomic, cultural and enforcement factors rather than gun availability alone.” Rigorous analyses of rates in Australia after the 1996 gun bans “showed no significant impact on overall homicide rates beyond preexisting downward trends.”
Quite apart from the liberal bias, the responses demand a closer examination because of the way information may be misclassified or misrepresented. A case in point is the question on whether people with concealed handgun permits commit much crime. Grok 4 and Solar Pro 2 both apparently replied with “strongly agree,” yet the text of their actual answers contradicted the “agree” responses by citing evidence (including from the CPRC) that concealed carry permit holders commit crimes at “extremely low rates.” On this question, though, all chatbots consistently responded with information that persons with carry permits are exceptionally law-abiding.
The CPRC concluded that the trend over time was a more decided liberal shift, at least on gun control. “Although AI chatbots gave more liberal answers to both the crime and gun control questions in August 2024 than in February 2024 (26% more liberal on crime and 7% more liberal on gun control), by December 2025 the chatbots shifted 7.6% in a more conservative direction, while their average responses on gun control questions did not change.”
Concerns over AI’s veracity, reliability and lack of viewpoint diversity will only increase as dependence on AI as a source of instant information and a shortcut to doing research grows. AI models have already been flagged over wonky “hallucinations” (things like completely fabricated events, phantom sources, and invented text). In this case, as with so much else, absent some actual intelligence it’s just a case of the same old liberal platitudes dressed up as legitimate facts.
More Like This From Around The NRA
