Small businesses rolling out AI agents to answer emails, chase invoices or manage support tickets may be handing them more system access than any human employee has, often without meaning to. That is the warning from identity platform JumpCloud, which says the tools most firms use to manage logins were never built for software that acts on its own.
AI agent access is running ahead of oversight at most organisations, according to new research from JumpCloud, which found non-human identities now outnumber human users in 83% of the businesses it surveyed. Some firms report at least six AI agents or automated accounts for every person on the payroll, yet only 21% have put any governance controls around them, according to JumpCloud’s account of its Q3 2026 IT Trends Report, published this week.
The pattern JumpCloud describes is a simple one: a team under deadline pressure gives an AI agent a broad, permanent login just to get it working, then never narrows that access once the job is done. “The population is growing,” JumpCloud says of these agents. “Oversight isn’t.” Each hardcoded token or standing credential left in place becomes another way into the business for anyone who finds it.
For a small business, the risk is less about scale and more about speed. A five-person firm that has wired an AI agent into its accounting system, CRM or inbox, in the way many are now doing with tools built on automation platforms like n8n, can end up with a single credential that reaches everywhere and belongs to no one in particular. Unlike a staff account, there is no leaver’s checklist to switch it off when the project ends.
JumpCloud is not alone in the identity space. Larger IAM (identity and access management) providers such as Okta and Microsoft Entra ID cover similar ground for bigger organisations, while smaller specialists including Teleport and Delinea focus specifically on managing privileged and machine credentials. JumpCloud’s pitch is aimed more squarely at smaller IT teams that want one dashboard for staff, devices and agents rather than several separate systems to keep patched and checked.
Its answer is a new tool, JumpCloud Workflows, a no-code automation builder that sits inside its existing platform. It lets an IT manager register an agent to a named owner at the point it is created, set how long its access should last, and have that access adjust or expire automatically rather than sitting untouched for months. JumpCloud describes it as one control plane for humans, devices and agents together, available now to its existing customers.
Any small business already thinking about writing an AI usage policy should treat agent credentials as part of that document, not an afterthought. The same logic applies to any tool connected via an MCP server, such as the setups covered in Smart SME’s look at Vestd’s MCP server for Claude and ChatGPT and Zoho’s expense management assistant: useful time-savers, but each one is another identity with a login that needs an owner and an expiry date.
The practical takeaway is straightforward. Before adding another AI agent, list what it can actually reach, give it the narrowest access that lets it do its job, and set a date to review or remove that access, exactly as you would for a temporary member of staff.
