AI AGENTS — PRODUCT DESIGN
Access control for AI agents and non-human identities
This is how Octoberus is designed for AI agents and services, and we discuss it with critical operators who want to apply it to their operation in a pilot. The problem it targets already exists: an agent that queries a historian, opens a ticket or asks an operations system for data usually does so with a person’s credential. With no identity of its own, no bounded permissions and no log that can be told apart from its creator’s, nobody can tell which actions were the agent’s and which were the person’s.
Non-human identities are already the majority. The controls are not.
109 machine identities per human one
According to the Identity Security Landscape 2026 report, there are 109 machine identities for every human identity. Each one is a key someone created and almost nobody reviews (source: helpnetsecurity.com, May 2026).
Only 34 % apply the same controls
According to Okta’s AI Agents at Work 2026 report, only 34 % of organisations apply to AI agents the same identity controls they apply to people. The other two thirds treat them as scripts (source: okta.com, 2026).
Identity abuse is already a top 10 risk
The OWASP Top 10 for Agentic Applications 2026 lists ASI03, Identity & Privilege Abuse. The CSA Agentic Trust Framework (February 2026) and the CISA-and-partners guidance on agentic AI in critical infrastructure (2026) point to the same control: identity and privilege per agent (source: genai.owasp.org, cloudsecurityalliance.org, industrialcyber.co).
An identity per agent, with a human owner
In the Octoberus design, each agent or service has its own identity type: name, purpose, the system that runs it and a responsible person, just like a contractor credential. ISO/IEC 27001:2022 A.5.16 and A.8.5 already cover identity management and secure authentication without asking whether a person is behind it.
Bounded permissions: least agency
In the design, each agent receives permissions per asset, per action and per time window, instead of inheriting the master key of whoever created it. It is least privilege, extended to the ability to act. NIST CSF 2.0 explicitly names “users, services, and hardware” in PR.AA-01 and PR.AA-03.
The same file as the doors
In the design, every action by an agent is logged where the door openings live: what it requested, under which identity, under which permit and what happened. One file to show the cybersecurity delegate, the CIP officer and the DPO.
PILOT — AI AGENTS
Let us talk about a pilot
Leave us your details and we will write to you to discuss applying this design to the agents and services in your operation.
Frequently asked questions
How does it relate to access for people? +
It is the same model: an identity with an owner, permissions bounded per asset and per time, revocation and an auditable log. In the Octoberus design, people, services and AI agents go through the same access gateway.
Can I see it applied to my operation? +
Let us talk about a pilot. We want to understand which agents already run in your operation, with which credentials and against which systems, to apply the permission model to your case. Leave your details below and we will write to you.
Where do I start? +
If you have agents or services reaching your systems today, the first step is to inventory them and give each one a human owner. For people, Octoberus provides permits per facility and zone with schedules, revocation and an exportable access log.