Why zero standing access is essential in the AI age

Imagine you're the head of security at a hotel. One day, a thief steals the janitor's uniform, including a ring of keys that can unlock every room in the building. In an instant, the thief has access to customer rooms and potentially your own office.

Criminals use a similar playbook to breach organizations by exploiting privileged access. As organizations deploy AI agents alongside human and machine identities, monitoring who can access what, and when, has become more important and complex.

Can't see the video? Watch on YouTube.

What is privileged access?

Privileged access management (PAM) is a cybersecurity category focused on protecting an organization's most sensitive systems, data, and infrastructure.

In the hotel example, that means the building keys and everything they unlock. In a business, the crown jewels might include a production environment, a database containing personally identifiable information (PII), or another customer-facing system.

"The crown jewels have changed over the years, but they've always been a combination of your more sensitive type of assets and the ones that are customer facing," Rom Carmel, VP for Product Management at 1Password, explained on the Random but Memorable podcast.

Privileged access covers more than who can reach a sensitive system. It also covers what that person, machine, or agent can do once they get there.

"Privileged access also refers to the actions you can perform," Carmel explains. "So it's a combination of high-risk actions plus the sensitivity of the target you could be performing those actions on."

"Privileged access also refers to the actions you can perform.”

What's standing access?

Return to the hotel. The head of maintenance always carries the ring of keys on their tool belt. That’s standing access: permission that remains active after a task, project, or role no longer requires it.

In your organization, standing access could mean permanent admin permissions, membership in a privileged group, or ongoing access to a sensitive app, cloud environment, or database.

AI-assisted coding tools have expanded the pool of people who can write and release code. At the same time, organizations increasingly host critical codebases in the cloud, where large, distributed teams can contribute. The more people who need privileged access, the greater the risk that elevated permissions will remain active long after a task ends.

"Essentially every engineer now has privileged access and has become a privileged user,” Carmel said. “We're now moving into a world where every employee to some extent can deploy code thanks to the vibe coding tools that are coming out. So every employee may become a privileged user."

“Every employee may become a privileged user."

Teams often know that privileged access is more widespread than it should be, but lack a complete view of who and what can reach their most important systems. Reviews become manual, fragmented, and difficult to verify. 

Reducing standing access requires a different operating model. Access should be granted when someone needs it, scoped to the task, and removed when the work is complete. This keeps access dynamic and responsive to people’s changing responsibilities. "Something that someone may need today is not what they may need tomorrow to support an incident or to be able to perform their work," Carmel explains.

Without standing permissions, an attacker who steals a credential will find it difficult to move laterally and reach critical systems without requesting additional access.

"Most attacks today start from existing access," Carmel said. "For example, imagine a phishing attempt that lets the attacker get into one of your employees' [accounts]. What they're able to do depends on the amount of access that person or device has. If you can get to a state where that person or device doesn't have any standing permissions, nothing is available to the attacker. That's when you're able to reach a zero standing privilege environment."

Just-in-time and just-enough access

Team members still need privileged access to perform legitimate work. But those permissions should match the task and expire when the work is done, in line with the principle of least privilege (PoLP).

Just-in-time access grants permission for the required period. Just-enough access limits that permission to the systems and actions the task requires. "Just-in-time and just-enough go hand in hand," Carmel explains. "Just-in-time access means that you're getting the permissions only to be able to support the time and the task that you need."

"Just-in-time and just-enough go hand in hand.”

When the task ends, the access should be revoked. A well-designed system should also record the request, approval, use, and revocation without relying on someone to clean things up later. 

Why AI increases the need for zero standing access

Businesses are rapidly adopting AI tools and workflows. A 1Password survey found that nearly half (46%) of developers are using AI agents in production environments today. 

While powerful, agents increase and complicate the risks associated with privileged access.

Agents can perform work at machine speed, which changes the scale of the access problem. You might be able to review a spreadsheet in five minutes; an agent can often do the same in seconds. "We can't necessarily review every action that an agent is taking because it's performing thousands of actions per minute," Carmel said.

But agents and humans also have something in common: they're non-deterministic. Unlike traditional machine identities, agents aren't 100 percent predictable. They can respond to changing inputs, select tools, and take different paths toward a goal. Sandboxes and guardrails help, but organizations still need to control their permissions and monitor their use.

An overprivileged agent therefore carries a substantial amount of risk.

"If you have an agent that can perform thousands of actions per minute, statistically it will use any standing access that exists in the environment," Carmel said. "And if it's overprivileged, it will use those overprivileges because it's just doing so many different tasks.”

Seventy-one percent of respondents said that AI agents have access to customer data, sensitive IP, or HR information at their companies. Meanwhile, 62% of all respondents reported security gaps in how their company managed agents. 

"If we want to give agents privileged access, we need to consider that they're a new type of entity that has the personality of or attributes of a human and also the attributes of a machine," Carmel said.

Zero standing privilege gives organizations a practical baseline. An agent requests access for a task. A human or automated policy evaluates the request, grants the appropriate scope, records how the agent used it, and revokes the permission when the task ends.

“It's important to get to zero standing access across your crown jewels and production environment,” Carmel said. “That's a baseline even to start to adopt agents securely. You have to go into a zero standing privilege environment so that you can do that safely."

Start with your crown jewels

Eliminating standing access isn't as simple as flipping a switch. Organizations need to understand where privileged access exists, who holds it, and how it is used.

Identify the systems and data that would create the most risk if an attacker reached them. Then map the humans, machines, service accounts, and AI agents that can access them. This often reveals dormant accounts, excessive permissions, and access that no longer matches a role or purpose.

Next, define how access decisions work. Who can request access? Which requests can policy approve automatically? Which requires human review? What actions can the identity perform, and how will you verify that access was removed?

For the hotel, this is the person responsible for the key safe. In a business, it’s something like 1Password Privileged Access that can deliver scoped access provisioned on request.

AI agents need the same discipline. You need to know which agents exist, what systems they can reach, and whether their actions match the reason access was granted. 

Reducing standing privilege replaces permanent permissions with access that matches the work, the identity, and the moment. Security teams gain control without slowing down employees and AI agents.