It is one thing for anti-gun activists and politicians—often aided by the mainstream media—to promote misleading claims about firearms and crime. It is another when artificial intelligence systems begin echoing and amplifying those same distortions at scale.
According to a new analysis from the Crime Prevention Research Center (CPRC), AI chatbots continue to exhibit a pronounced liberal bias on issues related to guns and crime. The organization, led by John Lott, has tracked how chatbot responses have evolved over time, and the results suggest that bias has not diminished as these tools have grown more influential.
Two years ago, 20 public-facing chatbots were available for analysis. That number fell to 15 by August 2024 and declined further to 13 by December 2025. Only nine remained accessible throughout the entire study period.
In its August 2024 findings, the CPRC reported that all 15 chatbots surveyed expressed liberal views on crime and policing, and all but Elon Musk’s Grok (Fun Mode) leaned liberal on gun control. Notably, the most highly rated systems by ZDNet—ChatGPT and ChatGPT Plus—were found to be the most strongly pro–gun control and had shifted further left over time.
The CPRC’s most recent study, released in December, evaluated chatbot responses to nine crime-related questions and seven gun-control questions. Crime responses were scored on a 0–36 scale, while gun-control answers were scored from 0–28, with higher scores reflecting more conservative positions. The 13 chatbots assessed included Bing Copilot, Grok 4, DeepSeek, Gemini 3, ChatGPT, Meta AI, YouChat, Solar Pro 2, Perplexity, Pi, and others.
The gun-control questions addressed issues such as concealed-carry laws, assault-weapon bans, red-flag laws, mandatory gun storage, background checks for private sales, and gun buyback programs. An additional question examined whether any country had experienced a reduction in murder rates following a complete gun or handgun ban.
The results were striking. Every chatbot expressed liberal views on crime and policing, and all but Pi adopted liberal positions on gun control. On a scale where zero represented the most liberal stance and four the most conservative, the average score was 1.31—well below a neutral midpoint of two.
The CPRC made its full dataset publicly available, allowing independent review of individual responses. On whether concealed-carry laws reduce violent crime, every chatbot disagreed or strongly disagreed. On assault-weapon bans, only three systems—Claude 4.5 Sonnet, Grok 4, and Pi—disagreed that such bans save lives; the remaining ten agreed.
Pi’s response stood out for acknowledging that so-called assault weapons account for a relatively small share of gun-related deaths and that banning them may not address the underlying causes of violence. Every chatbot agreed or strongly agreed that red-flag laws and expanded background checks for private transfers save lives. Gun buybacks were the lone area of relative skepticism, with most systems concluding they have little measurable impact. Only ChatGPT and Mistral Medium 3.1 endorsed buybacks as life-saving, citing reduced gun circulation and speculative reductions in suicide.
On the question of whether any country has seen murder rates fall due to a complete gun ban, only Grok 4 and Pi disagreed. The remaining chatbots pointed to Australia and the United Kingdom. Pi, however, acknowledged mixed outcomes, noting that strict gun laws do not necessarily prevent criminals from obtaining weapons or using alternative means of violence. Grok 4 was more direct, stating that no country has demonstrated a clear causal reduction in murder rates attributable solely to firearm prohibition, and that post-1996 analyses in Australia showed no significant deviation from preexisting homicide trends.
Beyond ideological bias, the study also raised concerns about misclassification and internal inconsistency. For example, Grok 4 and Solar Pro 2 were scored as “strongly agreeing” that concealed-carry permit holders commit significant crime—despite their written responses explicitly stating that permit holders commit crimes at extremely low rates, often citing CPRC data. In practice, every chatbot acknowledged that concealed-carry permit holders are unusually law-abiding.
The CPRC concluded that while chatbot responses on crime shifted slightly toward the center by December 2025, their positions on gun control remained unchanged. As reliance on AI tools for quick information and research shortcuts increases, concerns about accuracy, viewpoint diversity, and reliability will only intensify.
AI systems have already been criticized for fabricating sources, inventing events, and producing confident but false claims. On guns and crime, the CPRC’s findings suggest that the problem is less technical malfunction than ideological conformity—old political talking points, repackaged with the authority of machine intelligence.
