iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
At least two major enterprises outside the artificial-intelligence sector will suffer a critical AI-related security or privacy incident within the next 18 months, according to Michael Scott, chief information security officer at Immuta. Scott made the forecast in a Help Net Security interview published October 15, 2026. He attributes the risk to companies deploying high-risk autonomous agents for competitive advantage, productivity and cost reduction before the necessary safeguards are in place.
What Michael Scott is predicting
Scott’s statement is a forward-looking forecast, not a historical incident rate or survey result. He predicts that “at least two major enterprises outside the AI sector” will experience a critical AI-related security or privacy incident during the 18-month period following the interview.
The concern is the speed at which organizations are putting autonomous agents into operational environments. An agent may be able to retrieve data, call business APIs, make decisions or initiate actions without a person reviewing every step. When those capabilities are introduced before access controls and governance mature, a mistake or abuse can affect sensitive systems at enterprise scale.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Why AI agents create a different permission problem
Scott’s central principle is least privilege for non-human identities. He says organizations should “start treating agents as lower-trust identities,” rather than assuming that an agent is safe because it is acting on behalf of an employee.
#1 Best Overall
Agents should not automatically inherit a user’s access
The first audit question is which identity an agent operates under. An organization should know whether the agent has a dedicated identity with narrowly defined permissions or is using the requesting employee’s identity and access rights. Automatically passing a person’s complete data access to an agent can expose information that the task does not require.
Privilege creep can happen across people and systems
Human privilege creep usually develops as an employee accumulates permissions over time. Agents create additional pathways:
- A single agent may serve several employees whose permissions differ.
- The agent may call APIs or services that the human did not intentionally target.
- The agent may create or invoke sub-agents, extending the original authority into another execution path.
- Those actions can produce authorization decisions that were not anticipated when the agent was deployed.
These characteristics make a one-time permission review inadequate. The effective access of the agent must be evaluated in the context of each task, tool call and delegation step.
Rank #2
How organizations should limit agent permissions
Use task-specific least privilege
Give an agent only the data, tools and operations required for the current job. A request to summarize a specific project should not grant broad access to an employee’s entire mailbox, customer database or file repository. Permissions should be scoped to the task and removed or expire when the task ends.
Constrain sub-agents and delegated work
Delegation must not become a way to transfer unrestricted authority. Scott states that “an agent shouldn’t be able to hand off all its permissions just because it’s handing off a task.” A sub-agent should receive only the access needed for the delegated work, with that access ending when the work is complete.
Keep the authorization chain intact
Security teams need to reconstruct the path from the final action back to the original employee request. A service-account log showing that an action occurred does not explain who authorized it, what task prompted it or why the access was considered legitimate. Logs should therefore preserve the requesting user, the parent agent, each delegated agent, the tools called, the data accessed and the resulting action.
Rank #3
Why authentication alone does not secure an agent
An agent can be correctly authenticated and still have no legitimate reason to access particular sensitive information. Authentication answers who or what is presenting credentials; it does not establish whether the requested operation is appropriate for the task.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Scott draws a distinction between observing behavior and controlling it: “There’s a big difference between knowing what an agent did and being able to control what it’s allowed to do.” Buyers should ask vendors to demonstrate that unnecessary requests are blocked before data is returned or an action is executed, rather than merely recorded afterward.
Controls enterprises should require before deployment
Scott describes several safeguards based on Immuta’s own experience. Together, they form a practical governance baseline for organizations deploying employee-built or centrally developed agents.
Guardrails in the AI workspace
A multi-model AI workspace can include internally managed guardrails that apply consistent rules across models and agent workflows. The objective is to enforce data and action restrictions during execution, not just review an audit trail later.
Review before employee-built agents are published
Agents created by employees should pass a governance and security review before they become available to others. Reviewers can verify the agent’s identity, data sources, tools, delegated permissions, failure behavior and logging.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchApproval for new integrations
Connecting an agent to another application, database or API expands its possible impact. New integrations should require explicit approval so that the organization can assess the data exposed, operations enabled and identity used by the connection.
Continuing employee training
Training should continue after an agent is launched. Employees need to understand what agents can access, how to request changes safely, how to recognize unexpected behavior and when to report a suspected privacy or security issue.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.A practical review checklist for AI-agent programs
- Identify the runtime identity: document whether the agent uses a dedicated identity or inherited user permissions.
- Define the task boundary: list the exact data, tools and actions required, and deny everything outside that list.
- Set an expiry: make task permissions end when the task or delegation ends.
- Control delegation: pass only the minimum required authority to each sub-agent or external tool.
- Record authorization context: retain a traceable chain to the original requester, not only a service-account event.
- Test prevention: attempt unnecessary data requests and verify that controls block them before access occurs.
- Govern publication and integrations: review employee-built agents and approve every new system connection.
- Train and reassess: update employees and revisit permissions as workflows, models and tools change.
What the forecast means for enterprise security teams
The prediction highlights a governance gap rather than a claim that every autonomous agent is inherently unsafe. Risk rises when organizations treat an agent as a transparent extension of an employee, allow permissions to persist after a task, or rely on logging as a substitute for enforcement. Security teams can reduce that exposure by making non-human identities lower trust by default and requiring demonstrable, task-level controls before agents reach production.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
