Artificial intelligence (AI) preparedness isn't about developing a new security program. It's about ensuring the one you have is robust enough to address the challenges associated with AI.
Artificial Intelligence (AI) and cybersecurity preparedness are topics higher education institutions need to address, even as many stakeholders may be experiencing fatigue from ongoing discussions about AI. A great deal has been written and said about the impact of AI on the education and research mission. In security and IT offices, there is an equally urgent conversation that needs to occur about the impact of AI on the security program and on institutional risk. Specifically, how does this technology change the ability to protect the confidentiality, integrity, and availability of institutional information?
The data suggests that higher education institutions are behind on security strategy. EDUCAUSE found in January 2026 that 94 percent of higher education staff, faculty, and administrators had used AI tools for work in the previous six months, while only 54 percent were aware of any institutional policy governing that use.Footnote1 More than half were using tools that their institution did not provide.
Scenario
To paint a picture of the security challenges AI presents, consider this scenario. With human identities, security teams assign access based on a person's role at the institution. For privileged users, it is a combination of role and context (e.g., when the access is needed). But are IT and security teams considering how AI agents gain access to particular data stores, systems, and privileges? In the higher education context, many of these choices are being made in a distributed fashion and by the individual deploying the agent. The question a security practitioner should be asking is how "least privilege" and "need to know" are being maintained for agentic identities. What is keeping an agent from getting access beyond its prescribed scope or task? What controls are in place to ensure that no additional information is divulged?
The risk increases exponentially when a threat actor is able to manipulate an agent that may have more privileges than intended and access to sensitive data stores.
Foundations
The good news is that this is an evolution, not a revolution. The principles security teams already rely on—least privilege, need-to-know, and separation of duties—are the right tools for this problem, and they are already in the security team's toolbox. The challenge is that those principles were written for people, and they are now being applied to non-human identities and deployed by staff and faculty who do not report to IT leadership. So, AI preparedness is not about developing a whole new security program; instead, it is about whether the one that already exists is robust enough to cover this emerging area or requires some tweaks.
Getting there rests on two foundational elements that every good security program needs: governance and risk management. Three practice areas build on that foundation: good cyber hygiene, Zero Trust Architecture maturity, and AI-specific controls. None of the three work well without governance and risk management supporting them, which is why institutions should not address controls and architecture without first laying out a strong foundation.
Governance
Governance seems to be where many institutions are currently expending the most energy, and that focus is essential. It is important that those in charge of security have a seat at the table for the conversation.
First, structure is needed for the conversation. Leverage existing IT, security, and data governance groups if you have them; there is no need to reinvent the wheel. This is a listening opportunity more than a control point; no one needs security to take the air out of the room. Security and IT staff should understand how AI is being used and understand the business cases behind it because the deployments discussed in governance meetings are the ones that require creative protection approaches.
Next, policies should be reviewed with an AI lens to ensure that they hold up under this current shift. There was a strong conversation at the 2026 EDUCAUSE Cybersecurity and Privacy Professionals Conference, and the consensus in the room was that security teams do not need another policy. Instead, core policies should be robust enough to absorb a technology shift like AI.
Finally, security teams should inventory the AI tools in use at their institution, including where these tools are embedded in systems the institution already owns. This can be exceedingly difficult in decentralized environments, but inventory is the most concrete step in governance, and the one most institutions are missing. If the AI in their environment cannot be named, the rest of the security program is not covering it, regardless of what the institutional policies say.
Resource:
Risk Management
Governance builds the campus partnership, identifies the assets, and provides guardrails for the environment. Risk management zeros in on the threats and weaknesses for the institution and sets IT staff up for critical conversations with decision-makers about how to stay within risk tolerance.
AI threats run in two directions, and assessments should consider both. AI is being weaponized against higher education institutions, which raises the sophistication of phishing, social engineering, vulnerability discovery, and exploitation. Separately, the AI an institution has adopted is itself a target, and that is the direction most risk registers are missing entirely. An agent with standing access to a student information system is a target to be exploited. It belongs in the threat assessment alongside the system it can reach.
AI threats should be a part of routine intelligence gathering so they can inform risk assessments. It is worth running a scoped AI risk assessment against an established threat taxonomy. MITRE ATLAS, for example, catalogs adversarial techniques being used against AI systems. Scoping the assessment to AI keeps the focus tight and keeps the risk register from being overwhelmed. Regular assessment helps to inform leadership, build buy-in, and shape strategy, and it is the difference between knowing there is a coverage gap and being funded to fix it.
The three practice areas that follow all depend on the outputs of governance and risk management. If the foundational work isn't completed first, it may be difficult to bolster an institutional security program for AI.
Resources:
Good Cyber Hygiene
Good cyber hygiene still matters, and many institutions have struggled for years to get the basics done. AI does not change the basics.
Vulnerability management is where good cyber hygiene has an impact. AI has demonstrably compressed the timeline between a vulnerability existing and someone finding it, including in code that has been sitting in production for years.Footnote1 The window between disclosure and exploitation was already shrinking. Assume it will keep shrinking.
This compressed timeline has a direct consequence for a common practice in higher education. Some institutions have made a deliberate choice not to patch during the academic term. That was a defensible risk decision when exploitation took weeks to develop. It is a much harder decision to defend now, and it is worth revisiting explicitly rather than by default.
Code vulnerabilities are only one side of the coin. Configuration and misconfiguration, over-permissioned service accounts, forgotten integrations, and stale credentials are the other, and they matter more once agents are in the environment.
Zero Trust Architecture
Transitions to Zero Trust Architecture (ZTA) have been slow in higher education, but the platforms have matured, and adoption is easier than it was two years ago. AI raises the stakes on making progress.
Zero trust is exactly the model needed for an AI-enabled campus, where there is no certainty that an agent is acting in the best interest of the institution. The questions to ask about an agent are the ones Zero Trust was built for. In effect, Zero Trust says, "I don't trust you implicitly until I can validate and then regularly validate your request." In addition to identity, location, posture, and other contexts, purpose and scope should also be considered. When those conditions are addressed satisfactorily, least privilege and need-to-know have been applied to an identity that the policy was never originally designed to address.
For those who aren't sure where to begin with ZTA, a great place to start is understanding the current state of the institution's security and IT architecture. Security teams should start by assessing current Zero Trust maturity across the five domains: identity, endpoint, network, data, and application.
Resource:
AI-Specific Controls
Controls without governance and risk data behind them are just spending and chasing shiny objects. In today's budget-constrained environment, colleges and universities cannot afford to allocate funds without a strategy.
Return to the scenario at the beginning of this article: an agent with more privilege than intended, reachable by a threat actor, sitting in front of a sensitive data store. Open Web Application Security Project (OWASP) identifies both of the relevant failures: agent goal hijacking (ASI01), where an attacker redirects what the agent is trying to accomplish, and identity and privilege abuse (ASI03), where the agent's credentials become the attacker's access. Neither requires exotic tradecraft. Both exploit a failure to properly safeguard the agent.
Three control categories break that chain. The first is scoped credentials. An agent gets its own identity with permissions granted for its task, not the standing permissions of the human or service account that spawned it. That alone limits what a hijacked agent can reach. The second control category is human approval gates on privileged actions. If an agent can initiate a privileged action, something other than the agent should complete the action or authorize it. This is separation of duties and accountability to ensure an agent cannot take sole action on a critical business process. The third is ensuring security operations or the managed detection and response partner is aware of AI agents that are in use so they can monitor for anomalies. Agent activity needs to land in the same telemetry data analysts already watch.
This is one example of mitigating controls in AI and governance, and risk will help security teams prioritize.
The frameworks below provide more depth on controls than one article can. Start with OWASP for threats and mitigations, and use the Cloud Security Alliance AI Controls Matrix when control-level detail needs to be mapped to standards already used in reporting.
Resources:
Conclusion
So, is the security program ready for AI? For most institutions, the honest answer is not yet. The goal is to ensure the campus security program is robust enough to face this and future technology shifts. Security teams should build on what they already have and get serious about some things that have been sitting on the back burner for a while. Governance and risk management are foundational to effective security program development and are the right places to begin to develop the partnership and buy-in needed from campus stakeholders. A shared strategy can then drive prudent steps and investments that position the institution to better address the AI challenge.
Thank You
Notes
- Jenay Robert, The Impact of AI on Work in Higher Education, (EDUCAUSE, January 2026). Jump back to footnote 1 in the text.
- Eduard Kovacs, "Anthropic: Mythos Detected 23,000 Potential Vulnerabilities Across 1,000 OSS Projects," SecurityWeek, May 25, 2026 Jump back to footnote 2 in the text.
© 2026 Moran Technology Consulting.