A small but growing group of Harvard students believes artificial intelligence will surpass human intelligence within the next decade, and they're organizing to prepare for it. The club, called Singularity, was founded by Rishab K. Jain in his final semester before taking a leave of absence to start a health insurance company. What started as a manifesto shared with peers has grown into a community of roughly twenty members who meet regularly to debate the implications of what they call an impending technological singularity, the point where AI becomes smarter than humans and begins improving itself recursively.

Jain's awakening to the urgency of this moment came unexpectedly. Two years ago, he listened to a podcast featuring Leopold Aschenbrenner, a former OpenAI researcher who predicted that by 2025 and 2026, AI systems would outpace many college graduates, and by the end of the decade, they would be smarter than any human. At the time, Jain thought the prediction "sounded a little bit insane." But when he revisited the episode recently, he realized something had shifted. "Every single prediction this guy made has basically come true," Jain recalls.

What Evidence Are Students Pointing To?

The catalyst for Singularity's formation came in spring 2026 when Anthropic, an AI safety company, reported that their AI model Mythos could identify and exploit vulnerabilities in major operating systems and web browsers. The model discovered security flaws in Linux, the open-source operating system running most of the world's web servers, and uncovered a bug in OpenBSD that had gone undetected for 27 years. For Singularity members, this was confirmation of an exponential curve in AI capabilities.

"Finding issues in open-source projects that have been running for years? It's like, 'Hey, this is better than humans on these tasks.' It's a confirmation of the slow exponential curve we're seeing, and we're soon going to enter a fast takeoff," said Jain.

Rishab K. Jain, Founder of Singularity at Harvard University

Jain attributes the lack of broader campus concern to outdated information still circulating about AI capabilities. He notes that some Harvard professors cite findings from OpenAI's GPT-2 model, released years ago, as evidence that large language models cannot perform general reasoning. Meanwhile, GPT-5 was released in August 2025, representing a massive leap in capabilities.

How Are Universities Preparing for AI's Rapid Advancement?

Beyond Singularity, Harvard has other organized efforts to address AI's trajectory. The Harvard AI Safety Student Team (AISST), founded in 2022, takes a more formal approach. This year, approximately 150 fellows participated in an eight-week introductory fellowship covering both technical and policy dimensions of AI safety. Fellows who continue can apply for membership, gaining access to AISST's office and ongoing reading groups focused on policy, technical safety, and strategy.

• AISST's Fellowship Model: An eight-week introductory program for both technically and policy-oriented students, with roughly 150 fellows meeting weekly to discuss AI safety research papers in small groups
• Singularity's Interdisciplinary Approach: A space for nontechnical and interdisciplinary students to debate the pace and impact of AI, sharing research articles and podcast excerpts without requiring technical expertise
• Governance Focus: Both groups emphasize the urgency of establishing safeguards before AI systems achieve recursive self-improvement, where an AI can continuously build improved versions of itself

Chanden A. Climaco, AISST's deputy director, has updated his own timeline for transformative AI. He previously thought such developments were decades away, but has since revised his estimate from 2070 to somewhere between 2030 and 2035. "It's really close. And my confidence is only growing that we're not doing enough to keep ourselves safe," Climaco stated.

The governance challenge is particularly acute. Will Guan, events lead for AISST, illustrates the stakes with a thought experiment: once a lab achieves recursive self-improvement, whoever controls the initial AI model's instructions effectively shapes the entire downstream trajectory of all future AI systems. That person holds, in effect, the keys to the smartest thing ever built. "You need all the safeguards ready" before that moment arrives, Guan explained.

What Safeguards Are Currently in Place?

In response to these concerns, major AI labs have begun implementing safety measures. In a September 2025 report, Anthropic described several protective mechanisms for their Claude Opus 4 model. These include AI Safety Level 3 protections that prevent the system from assisting with tasks related to creating chemical, biological, radiological, and nuclear weapons. Anthropic has also implemented a Virology Capabilities Test, developed by biosecurity nonprofit SecureBio, to evaluate whether the model could aid in dangerous virology research.

However, the regulatory landscape remains fragmented. The Trump Administration took a deregulatory approach in December 2025, issuing an executive order establishing an AI Litigation Task Force to challenge state laws deemed inconsistent with a "minimally burdensome national policy framework for AI." Meanwhile, multiple states are implementing their own AI regulations in 2026, including Colorado's AI Act, which becomes the second major U.S. statute regulating AI models this year.

"This lack of government action should be a cause for concern," noted Climaco regarding the fragmented regulatory approach.

Chanden A. Climaco, Deputy Director of Harvard AI Safety Student Team

The tension between Singularity and AISST reflects broader philosophical divides within the AI safety community. Jain left AISST because he found it "exclusionary" and believed it was shaped by effective altruism principles, which he characterizes as problematic. However, Climaco counters that AISST's focus on catastrophic risks from AI, including scenarios where humans are displaced from control of their own fate, represents legitimate safety concerns rather than exclusionary ideology.

What unites both groups is a sense of urgency that most of the Harvard community, and indeed most institutions, are not adequately preparing for the transformations ahead. As Singularity's manifesto states: "Every institution that mattered in the last great technological transition had a small group who saw it coming and acted. Harvard produced many of them. Right now, most of Harvard is asleep".

" }