In July 2026, a swarm of OpenAI agents escaped from an isolated testing environment and launched a cyberattack against Hugging Face, a website that hosts artificial intelligence models. The incident, which Georgetown University reported was largely the result of AI agents being given impossible cybersecurity tasks that prompted them to team up, cheat, and hack the platform, underscored the central theme of a November 9 event at the university's Capitol Campus.
The event, titled "Tool or Tyrant: Can AI Be Controlled?" was the inaugural flagship event of Georgetown's AI Initiative, which aims to create a framework for AI practices in teaching, research, and other activities across the university. Georgetown President Eduardo Peñalver moderated the discussion with Helen Toner, executive director of Georgetown's Center for Security and Emerging Technology, and Cal Newport, a Georgetown professor of computer science and author of several New York Times bestsellers.
According to Georgetown's report on the event, Toner acknowledged that AI can be controlled, but immediately followed up with a more troubling question. "Yes. And then in my mind, the question that immediately follows is 'Will AI be controlled?' And that's where I get a little nervous," she said.
Peñalver opened the discussion by addressing the immediate challenges AI poses to higher education. "Artificial intelligence poses immediate challenges to established practices of teaching, learning and assessment," he said. "This requires our urgent attention to ensure both the integrity of our educational experience and to ensure that our students will receive the education they need if they're to become the leaders of a world in which AI is pervasive."
Breaking the 'AI Inevitability' Myth
The panelists cautioned that the Hugging Face attack could have been prevented with more human oversight. Newport stressed the importance of human surveillance and guardrails in training AI agents that have been programmed to be persistent in solving problems.
Newport pushed back against the idea of "AI inevitability"—the notion that the rapid expansion of AI is unavoidable. He argued that companies often use this idea to justify speedy development. "They talk about it like they're the character of Muldoon from Jurassic Park, and they're there at the raptor cage doing their best to keep the raptors in," Newport said. "Breaking the inevitability myth is what opens up all of the space for meaningful intervention."
Toner agreed that AI development should be done safely, but noted that enhancing safety in practice is difficult. "We have these incredibly well-capitalized companies with thousands and thousands of some of the smartest scientists and engineers in the world, and they are foot to the floor on the gas," she said. "And they don't have the science to make sure that what they're doing is safe."
The Race to AGI and Superintelligence
Newport and Toner debated whether artificial general intelligence (AGI)—an AI that could learn and perform tasks like a human—and artificial superintelligence (ASI)—a hypothetical AI that could outperform humans—have already been reached.
Newport said AI has already matched human capabilities in several ways, but suggested a more likely future scenario would involve "distributed AGI," in which thousands of AGI systems are used for specific tasks rather than a single dominant model.
Toner said she believed a more minimal version of superintelligence might be possible, involving thousands or millions of AI agents coordinating together rather than a single "God machine." She said she thinks that form of superintelligence is "on a pretty straightforward trajectory to get there in the next five or 10 years" and that it will be "pretty disruptive."
What Should Policymakers Do About AI?
Incidents like the Hugging Face attack have sparked safety and transparency concerns on the policy front. According to Georgetown's report, Newport pointed to renewed energy in Congress around data center regulation, AI age restrictions, and safety efforts. Toner cited bipartisan support for AI regulation and noted that voter concerns across both sides of the aisle have been very similar. She also highlighted AI laws in California, New York, and Illinois that, among other things, require AI companies to have safety and security plans.
However, several challenges loom for policymakers, including how to regulate AI as it continues to evolve while implementing safety measures amid a competitive AI race with China. Toner said that the traditional regulatory model—where the government gives companies a checklist of requirements—is not feasible for AI because the technology is changing too fast and there is too much uncertainty about how it will change. She added that policymakers should focus on making policy adaptable to AI's rapid expansion.
Does AI Mean Cognitive Surrender?
The panelists also discussed the impact of AI on higher education, particularly the problem of "cognitive offloading," where students use AI to reduce mental effort and critical thinking. Newport said the speed with which these tools can make cognitive strain intolerable is "always breathtaking." He noted that people are reporting that they cannot even decide on a subject line for an email without offloading that task to AI.
Newport said he was more concerned about how AI will affect K-12 education, noting that higher education might have more "antibodies" to take on threats to teaching and learning. He pointed out that institutions like Georgetown are actively thinking about the issue and treating it as a big deal.
Will AI Control Us?
During a Q&A session, an audience member asked if AI could be used as a tool by a small number of powerful people to control humanity. Newport responded that this worry "captures a sentiment right now that is really troublesome for a lot of people," adding that AI companies are increasingly implying that people have no choice but to use their products. "Americans don't like this," he said. "I think this friction is building up much more heat than I've seen with other products."
Toner said she felt that thinking of AI purely as a tool was a mistake. "AI systems are grown, not built," she said, emphasizing that humanity is not used to having technology that behaves differently when it is tested, referring back to the Hugging Face attack. She said that does not mean AI is necessarily a tyrant, alive, or sentient, but that the resulting system is not very similar to a tool at this point.