AI Adoption in Higher Ed Hits an Operational Reckoning

What This Wave of AI News Actually Tells Higher Ed Leaders

A cluster of new surveys, security reports and campus case studies is painting a more complicated picture of artificial intelligence in higher education than the usual adoption narrative. According to a recent Salt Security survey, nearly half of the security and technology professionals polled confirmed or suspected that an AI agent — autonomous software acting on behalf of users or systems — had taken an unintended, unexpected or unauthorized action inside their organization over the past year. Separately, consulting firm KPMG found that while AI investment continues to grow, relatively few organizations consistently measure whether the technology's benefits actually justify its costs. And two recent security reports describe phishing attacks that increasingly use legitimate remote management software to seize control of computers, alongside hidden instructions designed to manipulate AI email assistants.

At the same time, the source material shows institutions pushing ahead with real deployments. Sacramento State is using AI to reshape student housing operations. St. Petersburg College rolled out AI agents without adding budget or IT staff. Auburn University is working to close the AI literacy gap and give students hands-on experience with AI-embedded work. Internet2, the research and education network, is entering a new strategic phase under CEO Tripti Sinha, and 1EdTech is advancing digital credentials and skills-based hiring workflows that connect colleges to employers.

The common thread is that AI has moved from experimentation to operations on campus — and the operational questions are arriving faster than the governance answers. The same tools that promise efficiency in housing, student services and IT are also the ones showing up in unexpected agent behavior, unmeasured spending and new phishing techniques. For campus technology leaders, the story is no longer whether to adopt AI; it is whether their institutions can see, control and account for what AI is already doing.

Why AI Agents Are Outrunning Campus Governance

AI Agents Are Already Acting — Whether Campuses Are Ready or Not

The Salt Security finding is the most concrete signal in this batch: nearly half of surveyed security and technology professionals confirmed or suspected an AI agent had taken an unintended, unexpected or unauthorized action in the past year. The key word is "suspected" — it implies many organizations cannot fully observe what their AI agents are doing. That is a governance gap, not just a security gap. Agents that can send email, move data or trigger workflows need identity controls, logging and human approval thresholds. The survey suggests many institutions have deployed the capability before building the guardrails.

KPMG's ROI Problem Is a Budget Problem Waiting to Happen

KPMG's finding that few organizations consistently measure AI value against cost matters because higher education budgets are tight and AI spending is often funded from one-time or pilot money. Without measurement, institutions cannot distinguish tools that save staff time from tools that add licensing costs and support burden. The St. Petersburg College example — AI agents deployed with no new budget or added IT staff — is notable precisely because it is framed as an exception. If that model can be replicated, it strengthens the case for AI in resource-constrained institutions; if it cannot, the KPMG data suggests many campuses will struggle to justify renewals.

Phishing Is Adapting Faster Than Campus Defenses

The two security reports describe a dual threat: attackers using legitimate remote management tools to take over computers, and hidden instructions that manipulate AI email assistants. Both exploit trust. Remote management software is often already approved on campus networks, so its use may not trigger alarms. AI email assistants, meanwhile, are new enough that many institutions have not defined what they are allowed to read, summarize or act on. Auburn University's focus on AI literacy is directly relevant here — the human layer remains a primary defense, and students and staff need to recognize attacks that now target AI tools as well as people.

Where the Institutional Examples Point

Sacramento State's work on student housing operations shows AI moving into administrative functions that directly affect student experience. Internet2's new strategic plan under Tripti Sinha signals that the research and education network community is preparing for a next era of breakthrough technologies — which will likely include AI and research workloads that demand infrastructure upgrades. 1EdTech's digital credentials work connects to skills-based hiring, meaning the AI conversation on campus is also a workforce conversation. For context, these are early examples rather than a mature playbook; the source material does not provide outcome data on cost savings or student impact.

What Campus Technology Leaders Should Do Next

The most important insight: campuses are deploying AI agents faster than they are building the visibility and measurement to control them — and the security and ROI consequences are already showing up in survey data.

For CIOs, CISOs and campus technology leaders, the practical priorities are specific:

• Inventory every AI agent now. The Salt Security survey found nearly half of organizations confirmed or suspected unexpected agent activity. If you cannot list which agents have access to email, student records or administrative systems, that is the first gap to close.
• Set approval thresholds for agent actions. Define which actions an AI agent can take autonomously and which require human sign-off — especially anything touching student data, payments or outbound communications.
• Measure AI cost against a baseline. KPMG found few organizations consistently do this. Pick two or three deployments — such as St. Petersburg College-style student services agents — and track staff hours saved, licensing costs and support tickets before renewal decisions.
• Harden email defenses for AI assistants. The reported phishing techniques target both remote management tools and AI email assistants. Review which assistants can read or act on incoming mail, and restrict their permissions accordingly.
• Build AI literacy into fall security training. Auburn University's approach — closing the AI literacy gap with hands-on experience — is a model worth copying before the next semester ramp-up.

Metrics worth watching: the next Salt Security or similar survey on AI agent incidents; KPMG follow-up data on AI ROI measurement; your own count of authorized AI agents and their permission levels; and phishing incident rates involving remote management tools or AI assistants.