Einstein and Agentic AI: A Challenge for Higher Education

The emergence of an agentic artificial intelligence tool, dubbed Einstein, has sparked a significant reaction in the higher education sector this year. Einstein is designed to autonomously connect to Canvas, a learning management platform, to track courses, write papers, and submit assignments on behalf of students, all without the professors noticing.

This situation has highlighted a fundamental problem in higher education computing: the lack of reliable means to distinguish students from AI agents acting on their behalf on major learning management platforms. Josh Callahan, Chief Information Security Officer at California State University, stated, “The Einstein tool has been a real wake-up call. It underscores the fundamental challenge that all language model-based AI tools pose to higher education: how to assess student learning in a way that truly captures their knowledge and ability to apply that knowledge?”

While AI tools can enhance teaching and learning, they also pose a risk to human autonomy and creativity. These tools can potentially make important decisions or perform tasks that should contribute to student learning. Isaac Galvan, Director of Community Programs in Cybersecurity and Privacy for EDUCAUSE, asserts that higher education must “ensure that AI supports learning rather than replacing meaningful engagement in the educational experience.”

An Expanded Attack Surface

The threat of agentic AI is not limited to course management. It also extends to student portals, enrollment systems, financial aid platforms, and advising tools. Sandeep Kumbhat, Global VP and CTO for Okta, remarked, “AI goes beyond security, and higher education IT teams are struggling without adequate visibility into the tools being used.”

Higher education leaders realize that the issue of agentic AI is an identity security problem. “Agentic AI creates a challenge in identity and access management because it can blur the line between a human user and technology acting on their behalf,” explains Galvan.

To help address this issue, Josh Callahan suggests that educators require written essays and in-person exams, or video calls where students hold three fingers in front of their faces to reveal any facial overlay, known as the three-finger test.

Understanding Agentic AI as an Identity Security Problem

To mitigate identity theft by AI, institutions should implement stricter controls and verification processes to better validate users and their activities. Authentication controls can verify a student's identity, confirming that the student enrolled in a course is indeed the same person completing the work.

Higher education IT leaders must bolster identity security “by investing in identity and access management solutions that help verify real human presence,” asserts Galvan.

Additionally, behavioral analytics can detect agentic AI by tracking deviations from typical student behavior patterns on educational platforms. Instead of unreliable and error-prone AI detection software that focuses solely on student-generated content, network-level signals can help reveal AI impersonation by tracking technical metadata and login behavior to identify suspicious IP locations, for example.

Fighting fire with fire by using AI to enhance security is advised by Galvan. “AI can support predictive analysis of network patterns and improve monitoring and detection capabilities, allowing institutions to identify anomalies more quickly and accurately,” he says.

Building Effective Governance Frameworks

Before rushing to adopt tools, higher education institutions must first establish an effective AI governance framework. Institutions should create a general steering committee on AI that brings together various stakeholder groups to determine the institutional strategy regarding AI, advises Josh Callahan. “That’s the first thing to do. It always comes back to people, processes, and tools — in that order.”

Given that rapidly evolving AI systems rely on multiple interconnected data processes and are increasingly integrated into software platforms, institutions must adopt a collaborative governance approach that involves colleagues from across the organization in decision-making and oversight, explains Galvan.

“Schools should consider agentic governance as an extension of their governance of human identities,” says Sandeep Kumbhat. In practice, he explains, such governance may involve certification campaigns synchronized with semesters and enrollment cycles rather than annual exams, access with the principle of least privilege limited to specific tasks, and ongoing discovery to detect phantom agents acting outside of core IT.

However, Galvan warns against overly rigid governance. “Governance frameworks must strike a balance between caution and innovation,” he says. “Overly restrictive policies can breed resentment, stifle innovation, or be ignored by stakeholders eager to adopt new technologies.”