Loyola Responds to AI Safety Concerns

On Sept. 9, ex-Anthropic researcher Jacob Coxon gave an exclusive interview with The Wall Street Journal where he voiced safety concerns about the development of a fully recursive self-improvement artificial intelligence model. 

The warning sparked a chain of industry concerns on the development speed of AI and safety concerns ensuring artificial intelligence doesn’t “spiral out of human control,” according to NPR. Once fully developed, a recursive self-improvement AI model would be able to rewrite its own code to correct itself and “evolve” without any human intervention.

Regardless of the possible benefits, Coxon said his main concern was AI’s developmental speed and the possibility it would soon be increasingly difficult for humans to control AI and its capabilities. 

Loyola’s faculty experts in AI and social justice, Abha Rai and Jonathan Singer, voiced concerns about bias in the data AI is trained on and the social implications of allowing AI to self-develop without any human oversight or intervention.

“AI isn’t trained on clean data. It pulls from sites like Reddit, YouTube, and whatever books it can gain access to — all which are susceptible to racial and social biases,” Singer said. “It’s an incomplete data corpus that isn’t at all representative of human thought.” 

Singer considered the advancements AI could bring, but he reasserted that as social workers, he and Rai are conscious of the fact that what may be good on a national level isn’t always good for the individual or community.

Rai said she had a similar outlook, but suggested the issue isn’t with the evolution of a fully recursive self-improvement model, but human accountability.

“AI can’t replicate human thought and ethics,” Rai said. “We must have human intervention and bias checks. The evolution of AI could be a problem, but the main problem is how we facilitate human usage of AI.” 

In the event of a public release of a self-improvement AI model, Rai said there’d be a need for a conduit to facilitate accountable use of AI. For Rai, that conduit would be social workers. 

“Social workers have a large responsibility to provide access to and education around AI, but ensure its benefits to the people offset the costs,” she said. “Problems arise with AI development only without human connections and intelligence.”

Like Rai, Associate Professor of the Computer Science Department Mohammad Abuhamad said the central concern around a recursive self-learning AI model shouldn’t be its evolution or capabilities, but what regulations and rules are in place for its usage.

“The benefits coming from AI are undeniably great and have brought a lot of competitive advantage in many domains and industries, but we have rules for everything, and AI should be a part of that,” Abuhamad said. “Our regulations should be based on what we allow people to have AI generate, around the implementations and limitations of use.”

Abuhamad said it’s not the responsibility of the developers or those in the AI fields to determine the limits, but instead of sociologists, government workers and of people who know best what people need.

“In my opinion, the developers in the AI field have no place making or deciding the regulations and rules because it’s a policy issue, not an AI issue, and should be dealt with as such,” Abuhamad said. “When it comes to making rules for use, we should want to put our people and community first.”

The current federal administration has one law around AI usage, known as the TAKE IT DOWN Act, but no general comprehensive regulation like the European Union’s AI Act. In 2025, President DonaldTrump signed an executive order to block any state regulations meant to reign in or regulate AI.

Trump dismissed any concerns regarding industry warnings around the speed of AI development and the current lack of regulation. He said all the United States needs is a “strong and smart” president,  the Associated Press reported. 

Fourth-year computer science major Beatriz Santos said she’s particularly concerned with the level of trust people seem to have in AI and suggested a recursive self-learning model could make it worse.

She pointed to Elon Musk’s SpaceXAI model, Grok, which is marketed as a “truth-seeking AI” frequently used as both a “fact-checking” tool and a conversationalist on the social media platform X. However, Santos said this is concerning because there’s no guardrails around the type and volume of data AI pulls from to give a “truthful” response.

Nevertheless, Santos said people shouldn’t shy away from learning about and using AI. 

“There’s a lot of fear mongering, but having understanding around AI can elicit more agency and productive conversation,” Santos said.

For some students like first-year biomedical engineering major Rahul Rajeshkumar, it’s more difficult to remain optimistic. He said he switched from studying computer science to biomedical engineering after learning more about AI’s rapid development. 

“I wanted to pursue a career that was less likely to be taken over by AI,” Rajeshkumar said. “I guess I wanted more career security.”

Rajeshkumar said he’s concerned with how people somewhat place complete trust in AI and pushed back against the idea of a fully recursive self-learning model.

“It wouldn’t be good for the public,” he said. “We already have such a big problem with media literacy and people relying on AI that something as close to sentience as a self-learning model would probably be extremely problematic.” 

This idea was contested by fourth-year neuroscience major and AI Society Club President Cordelia De La Fuente. Although she said AI shouldn’t be making decisions in place of people, she argued AI could open new doors and serve as a tremendous equalizer in the medical field between developed and less-developed countries.

De La Fuente believes even in light of a recursive self-learning model, people will prioritize human power and control over innovation.

“Without the safety of people confirmed, regulation of innovations will always make sense for people,” De La Fuente said. “In the end it’s not artificial intelligence that we need to worry about, it’s natural intelligence.”

Though the perspectives and experiences with AI varied throughout these conversations, many held a consistent message. As iterated by both De La Fuente and Santos, the well-being of humans should be above any profit, innovation, or corporate success. 

As the United States continues to debate how to regulate AI and remain in control, Abuhamad said humans should keep in mind how to best engage with ever-advancing technology.

“The message is simple, the problem is not about the evolution of AI, but how we use it.”