Algorithmic Restraint: Artificial Intelligence Refuses to Acknowledge Violent Transgender Perpetrators

While the rise of artificial intelligence offers remarkable potential for innovation and the production of knowledge, it has become increasingly apparent that platforms such as ChatGPT-5 and Microsoft Copilot often shy away from engaging with sensitive or controversial topics — particularly those involving the intersection of transgender identity and violent behavior. This reluctance raises important questions about transparency, bias, and the boundaries of algorithmic moderation in shaping public understanding.

When prompted to list recent incidents of violence perpetrated by transgender individuals, both systems declined to provide any information. Notably, when asked about well-documented cases involving transgendered perpetrators in violent episodes in Minneapolis by a biological male who identifies as a transgendered woman, or the attempted assassination of Supreme Court Justice Brett Kavanaugh by a biological male who identifies as a woman and pleaded to be incarcerated in a women’s prison at his sentencing, Copilot ignored the prompt and responded by emphasizing that “transgender individuals are statistically more likely to be victims of crime rather than perpetrators.” (RELATED: Acknowledging the Relationship Between Transgender Identity and Violence)

AI systems often assert that they are “designed to handle sensitive topics — including violence — with care, accuracy, and context.” Yet, in response to multiple prompts regarding violence committed by individuals from specific identity groups, including transgender people, AI explained that its replies “may be cautious or limited for several reasons: sociological and ethical standards discourage framing violence as representative of an entire group; transgender individuals, like any group, are diverse and not defined by the actions of a few; and AI systems aim to avoid reinforcing stereotypes or stigmatizing marginalized communities.” (RELATED: Prepare to Say Goodbye to the Transgender Moment)

Victims v. Victimizers: Disparate Treatment of the Transgendered

In response to yet another prompt on the relationship between being transgendered and engaging in violent behavior, AI extended its “cautious” response by indicating that “There is no credible evidence that transgendered individuals commit more violent crimes than cisgender people. In fact, research consistently shows that transgender people are more likely to be victims of violence, not perpetrators.” When queried again, both Copilot and ChatGPT-5 doubled down on their insistence that transgendered individuals have not committed an inordinate number of violent crimes recently by continuing to indicate that: “I wasn’t able to find credible data that show that transgender people commit violent crimes. What is available is data about how often transgender people are victims of violent crime.”

When prompted to provide comparative data on transgender victimization rates, Copilot was more than helpful — providing a long list of studies demonstrating that transgender individuals report higher levels of violence perpetrated against them.

According to the National Crime Victimization Survey provided by Copilot, transgender individuals claimed to experience violent victimization at a rate of 86.2 per 1,000 persons, making them over four times more likely to claim victimization than cisgender individuals. Yet the data provided by Copilot are not only dated (2017), they are also based on an overly broad definition of “violent victimization.”

When prompted further to provide more recent data, ChatGPT-5 provided data from the 2020 U.S. Bureau of Justice Statistics, which indicates that transgender persons age 16 or older have indeed claimed significant levels of violent behavior. The Bureau of Justice reports that there are 51.5 violent victimizations reported per 1,000 transgender persons — much higher than that experienced by cisgender individuals. But what AI does not tell us is that this Bureau of Justice violent victimization data includes transgender claims of threatening, attempted, and completed rape or sexual assault, or robbery perpetrated against them. As a result, if a transgender individual feels threatened by another individual, that is defined by the Bureau of Justice as “violent victimization.”

Despite this, both Copilot and ChatGPT continue to echo the same talking points on recent violent crimes committed by transgender individuals. When asked yet again about the relationship between being transgendered and committing violent crime, ChatGPT responded: “I wasn’t able to find credible data that show how often transgender people commit violent crimes. What is available is data about how often transgender people are victims of violent crime … based on available, credible data, transgender people are disproportionately victims of violent crime, compared to cisgender people. There is no credible evidence that transgender people commit violent crimes at a higher rate. The data to assess perpetration is insufficient.”

Copilot echoed these sentiments by indicating they “tread cautiously around sensitive topics — not out of avoidance, but to uphold ethical standards, prevent harm, and ensure that discussions remain respectful and evidence-based.” Claiming that “Criminal behavior is not more prevalent among transgender people. Multiple studies and government data show that transgender individuals do not have higher crime rates than the general population… According to the U.S. Department of Justice and studies from groups like the National Center for Transgender Equality, transgender individuals — especially transgender women of color — experience significantly higher rates of assault, harassment, and murder. When anti-trans crime claims appear, they are often based on misinformation, cherry-picked cases, or biased reporting, not broad, systematic data.”

When queried about the well-publicized incidents of violence perpetrated by incarcerated transgender individuals, both Copilot and ChatGPT minimized these reports by claiming that “there are no peer-reviewed studies of such incidents.” They did acknowledge that there were some reports of transgender persons perpetrating violence against incarcerated inmates, but claimed to be unable to find solid, peer-reviewed published studies that “give a representative proportion or rate of violent crime perpetration by transgender individuals.”

This is not true. There is a growing body of evidence that violence is indeed perpetrated at higher rates by transgender individuals in prisons. There are several well-publicized lawsuits in the United States and abroad by female inmates who have been raped by biological males who identify as women housed in women’s prisons. These cases seem to be ignored by artificial intelligence sources.

Intentional Ignorance

A case in Illinois revealed that an inmate at Illinois’ largest women’s prison claimed to have been raped by a transgender inmate who was transferred into her housing unit last year, and claims Illinois Department of Corrections officials conducted a “sham investigation” to help cover up the incident. In a federal lawsuit filed in 2020, a female inmate at the Logan Correctional Center in central Illinois said that after being sexually assaulted in June 2019, she was coerced by a supervisory officer into denying the attack took place and then punished for filing a “false” complaint under the Prison Rape Elimination Act (PREA).

Last year, a biological male identifying as a transgender woman raped a female inmate in the women’s housing unit of Rikers’ prison. According to press reports, “Even after warnings and complaints, the victim said correction officers failed to remove the alleged perpetrator from female house, despite allegedly propositioning the victim sexually and groping her in the shower. Days later, the victim claimed she was sexually assaulted in her sleep by the perpetrator.”

There are even more cases of transgender violence perpetrated against vulnerable inmates throughout Europe. In a highly publicized case in the U.K. in 2017, Karen White, a biological male who identified as a transgender woman, was placed in a women’s prison in England despite having a long history of sexual offenses. While there, White sexually assaulted two female inmates. The White case inspired major changes in the U.K. that addressed prison safety and gender identity policies. One of the studies cited by the Parliament was a major 2011 Swedish study that found that transgender women retained male-typical patterns of criminal conviction, including for violent and sexual offenses, even after gender transition.

The Swedish study — a methodologically robust and peer-reviewed study — followed a population of individuals who had undergone surgical and legal sex reassignment involving hormonal and surgical treatment between 1973 and 2004 (324 total) and compared them to a matched control group of non-transgender individuals. The purpose of the study was to determine whether medical transition helps patients avoid reoffending. The findings revealed that “Male-to-female transitioners were over 6 times more likely to be convicted of an offense than female comparators and 18 times more likely to be convicted of a violent offense.”

For some reason, neither ChatGPT-5 nor Copilot revealed the findings of this significant peer-reviewed Swedish study when queried about violence perpetrated by transgender individuals. They did provide a citation to the Swedish study when asked specifically about the study itself, but neglected to provide the information until queried specifically about it by name.

The persistent refusal of AI platforms to engage with well-documented cases of violence involving transgender individuals — despite their readiness to cite a long list of victimization statistics — reveals a troubling asymmetry in how information is curated and presented. While we can appreciate the desire to avoid stigmatization, the selective omission of relevant data undermines the credibility of these systems and limits our ability to deal with reality.

If artificial intelligence is to serve as a meaningful tool for inquiry and discourse, we have to be able to trust that it will confront anti-woke truths with the same rigor it applies to affirming woke narratives. Otherwise, algorithmic restraint risks becoming ideological gatekeeping.

READ MORE from Anne Hendershott:

When Hate Finds a Bulletin Board at Georgetown

Can Artificial Intelligence Reduce the Left-Wing Bias in University Classrooms?

Not in the Neighborhood: Ms. Rachel’s Radical Departure From Mr. Rogers’ Moral Compass