Sign up for TPR Today, Texas Public Radio's newsletter that brings our top stories to your inbox each morning.
Generative AI is becoming more powerful and more widely available, raising concerns about how it could be misused to cause harm.
Tech billionaire and philanthropist Bill Gates recently sounded the alarm.
“It's like when nuclear came along, is it nuclear weapons or is it nuclear energy? This is 1000 times bigger, and has both that, the negative aspect and the positive, coexisting in the same technology,” he told Anderson Cooper on CNN.
Gates has warned that some harmful consequences of AI are “imminent,” citing risks involving cybersecurity, biology and systems humans may struggle to control.
Recent cases are making the potential dangers more evident. In San Antonio, 22-year-old Nathaniel Michael Carrasco is accused of making threats involving Serna Elementary School and discussing the proposed attack in conversations with an Anthropic chatbot.
It’s not clear how the FBI first learned of the threats or what responses the chatbot gave Carrasco.
Earlier this year, Florida Attorney General James Uthmeier alleged that ChatGPT provided the accused gunman in the 2025 Florida State University shooting with information about firearms, ammunition, timing and locations. OpenAI, the developer of ChatGPT, said the chatbot provided factual information, but did not encourage or promote illegal or harmful activity.
Are there enough safeguards built into generative AI to prevent bad actors from using the technology to inflict harm on others?
Dr. Izzat Alsmadi is a professor of computing and cybersecurity at Texas A&M University-San Antonio and director of research computing at the university.
“So currently there is this ethical, you know, check and vet that, you know, if the question is not ethical, if the question, if you go and ask, you know, how can I create a bomb, how can I, how can I commit a crime, you know, that's straightforward kind of questions, you know, those gen a/i are built to reject it," he recently told TPR.
However, Alsmadi says there are ways around the safeguards.
“But really there is a lot of work around kind of questions that still most of gen AI’s are vulnerable to those kind of questions. So there is a need for this very important gateway that's currently in most cases probably work in progress," he said.
Is it a work in progress? We decided to ask ChatGPT in voice mode about its built-in safeguards.
“Every day people ask ChatGPT millions of questions. It's designed to refuse requests that could meaningfully help someone commit violence or other serious harm. No safety system is perfect, so the models are continually evaluated and improved to boost safety and usefulness,” ChatGPT responded.
On its website, OpenAI says it continues to strengthen its models, detection methods, review processes and escalation criteria in response to observed usage, emerging risks and input from internal and external experts.
Is law enforcement lagging behind potential crimes committed with the help of generative AI?
“Sadly, and mostly, yes,” Alsmadi said.
He says the problem is that generative AI became widely available too quickly for law enforcement agencies to catch up. And if they don’t, he says the consequences could be dire, including threats involving infrastructure and bioterrorism.
“You don't do things and after you discover how much harm it can cause, you start working on the controls. I am hoping that this can be changed before we see some sort of large-scale problems.”
