For MSPs, AI readiness starts with knowing who – or what – has privileged access and who’s accountable for it.
Imagine an MSP connecting a new AI agent platform or series of homegrown AI agents to its core systems, SaaS applications, and PSA platform. The goal sounds straightforward: automate manual tasks, summarize tickets, retrieve account details, or more sensitive tasks like updating customer data. The goal is for MSP technicians to focus on higher-value customer satisfaction efforts and spend less time on lower-level administration and configuration.
Then come the questions about privileged access and authorization. What permissions should agents have? Should the agent see every customer account? Can it change contact information or make updates without approval? How is this controlled at scale? Who can stop one or more agents if something goes wrong? Who is ultimately accountable for all agents, their positive actions, and their mistakes?
Today, MSPs and their customers are rushing to deploy professionally built AI solutions and homegrown AI agents with minimal consideration for the long-term effects of proper guardrails for identity and privileged access management. We believe a mindset shift is needed to prevent unintended breaches or non-malicious configuration changes that could cause system failures.
Identity and Privileged Access Management has always been part of running a business – who can log in, who has access to data, what an administrator can and cannot do, and how admin privileges are escalated and scoped. What’s changing is that agents are doing these jobs too, and their identities now outnumber human identities by a factor of 100:1. We’re still in the early innings of agentic assistance, so this ratio is expected to increase.
Now, every employee and AI agent is a digital identity capable of acting on its own. As AI agents and agentic software solutions expand their scope and control across more applications and systems, decisions governing their access, authorization, and authentication are an even more important component of how MSPs and their customers build reliable, trusted infrastructure.
That’s the mindset shift MSPs need to get ahead of – how can we as MSPs become the AI authority and heroes for both human and non-human identities (“NHI”) to adopt AI safely and with confidence. If they don’t have proper privileged access and identity in place for humans pre-AI deployment, the problem is only exacerbated post-deployment of AI agents. That is real liability for the MSP, and it’s amplified when extrapolated across tens or hundreds of customers.
AI readiness includes privileged access readiness
The good news is that MSP customers already anticipate needing help here and MSPs themselves still have time to get a managerial plan in place. In Barracuda’s MSP Customer Insight Report 2025, 39% of organizations said they expect to need MSP support with AI tools and AI-driven applications over the next few years. That means MSPs have significant opportunities to become the AI voice of reason with their customers, and it should start with a conversation about AI readiness.
Selecting the correct AI agent-based tool, or building your own agent or agents, is only one decision process. MSPs and their customers also need to evaluate how much authority they’re comfortable assigning to each agent. An agent that summarizes emails and sends those summaries to employees via Teams, for example, carries a somewhat different risk profile than an agent that reads and updates databases with sensitive records.
NIST’s National Cybersecurity Center of Excellence raised this exact concern in a February 2026 concept paper, warning that giving AI agents access to data, tools, and applications creates risks that existing identity controls weren’t built to handle. Public comment has since closed, but MSPs don’t need to wait for a federal standard to start asking the right questions and architecting their own businesses, and their customers’ businesses, correctly.
Existing gaps are amplified with AI running on top
Consider the ultimate example of an AI agent assisting an MSP technician with daily tasks and efforts. Let’s call it the “technician agent.” If a “technician agent” is built to log in to MSP systems or MSP customer systems, emulating the human admin, and conduct work on behalf of that human technician automatically, what is the agent allowed to do specifically? Alternatively, what if the agent is assigned service account authority from an account with broad access granted years ago, and that access isn’t well controlled or defined? Neither case requires an attacker per se, but both cases create risk of catastrophic configuration mistakes if the agent is not controlled and assigned zero standing privilege or least privilege.
Agents for MSP customer users get murky the same way. Let’s assume the customer employee authorizes an agent to perform work on their behalf, then changes departments six months later; however, the agent continues to have full rights and privileges. Without a named successor and a review date, nobody is monitoring whether that agent and its access level(s) still make sense. This should be defined by the MSP with their customer long before the agent is deployed and activated.
These are small examples of how serious AI agent privileges and controls could become, but again, we must think about this at exponential scale – tens of thousands of MSPs managing millions of AI agents for over a million of SMBs globally. That is why Identity and Privileged Access Management could be viewed as critical infrastructure – excessive permissions, orphaned accounts, agents with no clear owner, no primary agent control plane, and general lack of accountability.
It’s time for MSPs to become the AI authority for their customers and differentiate themselves with proper readiness attention, conversations, and soon – the right MSP-focused privileged access control plane. That is part of the real opportunity for MSPs as we enter this new world.
Ask yourself and your customers five questions before deploying agents:
1. Who owns agent actions?
MSPs should agree on an agent’s purpose on behalf of their customers and assign ownership responsibilities to technicians. Review agents and their access whenever a workflow or human org chart change occurs. Assuming this will all just “work out” is inadequate.
2. Whose authority does it use?
Make sure to note whether each agent is acting on behalf of a technician or an end user, or based on service permissions – it may be more than one. Also, check if that authority is broader than what the task requires.
3. What can the agent actually do?
Set clear permissions focused on the specific tasks needed, such as reading ticket queues and adding service notes for the technician, rather than granting multiple capabilities, such as deleting records or exporting the database. Also, once the task is completed, the agent should go dormant rather than remain active with standing privileges. Any credentials the agent has should be revoked.
4. Which actions need human approval?
Make sure to build an approval framework and agree on when a human should be involved. Incorporate that approval process into the agent deployment and monitoring workflow so it happens automatically, keeping authorization operations efficient.
5. Can we trace its actions and cut off its access?
Confirm that AI agent activity appears in logs, and show which agent performed an action, what changed, and when. Test the process for shutting down an agent and revoking access, and confirm that the process is durable.
Turn this into a service, not a one-time favor
These five questions work best as a repeatable, managed offering, not a conversation given away for free. Package it as an AI Access Readiness Review – a fixed-scope engagement that documents AI agent and non-human identity purpose(s), connections, owners, permissions, and approval requirements, then identifies the highest risk gaps. Ideally, this occurs before agents are deployed, but it’s still relevant after deployment.
Assign clear responsibility: the customer owns the decision about which agents it needs to run its business, but MSPs can advise on best security practices and help verify agents’ capabilities, controls, and permissions. During quarterly business reviews, offer an AI Agent Access Review that customers can actually understand and demonstrate monitoring capabilities and wins.
As identity and privileged access management controls for AI agents and non-human identities become a larger and more integral part of the attack surface, MSPs need a more consistent way to understand and manage risk across customer environments. For MSPs – at your next customer meeting, pick one AI agent and work through these five questions with key stakeholders at your customer. Their answers will help guide you on how to best craft your AI Readiness offering, demonstrate how you can add value, and establish your managed service company as an authoritative expert and a trusted partner.
Michael Roth is Vice President of Identity Product Management at Barracuda Networks, where he leads the expansion of identity security capabilities for partners and MSPs. Previously the Founder and CEO of Evo Security, 
Michael brings deep expertise in identity management (IdM), privileged access management (PAM), and MSP-focused security solutions. He is passionate about helping partners strengthen cyber resilience through modern, scalable identity protection.
