What is an Agentic ISMS, and what stays with people?
An Agentic ISMS is an information security management system in which AI agents take part in the work: they derive risks, measures, controls and policies from an organisation's own documents, fit new requirements into what already exists and carry out tasks. A person sets how far an agent may act on its own, and every step remains traceable.
This is the term we use for OdySecure, the Agentic ISMS by Cybervize. The definition is also on our Agentic ISMS page, together with a comparison and answers to common questions.
Why the term is coming up now
AI systems are called agentic when they work towards a goal and choose for themselves which step and which tool to take next. They read documents, call tools and enter results into systems themselves. Systems like these are now arriving in business software. GRC vendors market "agentic GRC", and vendors of IT service software market "agentic ITSM". Agentic GRC usually refers to tools in which AI agents handle individual compliance tasks, such as collecting evidence or filling in questionnaires.
In an Agentic ISMS, the agents work inside the management system itself. They build the model of risks, measures and controls and fit new frameworks into it. In the security processes, they take on tasks up to incident triage.
What changes compared with classic ISMS software
Classic ISMS software runs the management system with its assets, risks, measures, policies and evidence. In the usual set-up, entries come from questionnaires, catalogues and templates that someone in-house works through.
In an Agentic ISMS, the AI does this groundwork. Suppose an organisation uploads its network diagram, the contracts with two service providers and its emergency manual to OdySecure, and adds details about its structure. The derivation extracts the text, recognises the infrastructure and derives risks, measures, controls and draft policies from it.
The result is a set of proposals. The information security officer accepts them one by one or in bulk, and rejects what doesn't fit. The derivation creates policies as drafts. They apply only once they have been reviewed and approved.
The officer's work shifts from creating entries to reviewing them. OdySecure records the reasoning for the proposals and their supporting passages in the documents. Whether a risk is right for the organisation and a measure is adequate is still judged by someone who knows the business.
Three ways agents work in an ISMS
Agents take work off your hands in three ways: they derive, they answer and carry out instructions, and they act under their own identity. We explained why this order matters in Agents are the last step, not the first.
Derive
The example above shows the first mode. It also covers fitting a new framework into what already exists. The AI maps the framework's requirements to the existing risks and checks which of them existing measures already cover. For the rest, it proposes measures. Where a gap raises a new risk, it adds that risk as an unassessed proposal. The risk owner assesses it and decides whether it stands. This mode is available in OdySecure. It builds the register the other two modes work on.
Ask and instruct
Ask your ISMS. Every answer comes with its source. In OdySecure, the Navigator does this. It answers a question like "How many open measures at the northern site are past their deadline?" from your organisation's data. OdySecure attaches the sources from the data it queried, not the language model. If the Navigator finds nothing to base an answer on, it asks you to clarify. OdySecure logs your questions in your organisation's audit log.
On your instruction, the Navigator also does work in the system. If you write "Create a task to review the password policy" in the chat, it first shows you a plan with the entries it intends to create. Nothing runs until you have confirmed its plan, and it only acts with your permissions. In the same way, it links measures to risks or records KPI values. It files a risk acceptance or a time-limited exception as a request for the responsible people to decide on. You can usually undo your last five confirmed actions within two hours. The Navigator is available in OdySecure.
Act
In the third mode, connected agents carry out tasks themselves, for example incident triage. In OdySecure, the status is: design-partner programme, not yet generally available. An agent there works under its own identity and with its own role. A person sets its autonomy level for each task. Proposing is the default. Letting an agent act takes approval from two different people. An agent can never close an incident, accept a risk or grant itself permissions. A kill switch locks an agent out and revokes its access. If you want to use this early, you can become a design partner.
How OdySecure secures the agents
An uploaded document can contain hidden instructions, and a language model can be wrong. That is why OdySecure enforces its limits outside the language model. The system's permission check blocks what an agent may not do, whatever the prompt says. The server carries out a Navigator change only after you have confirmed it. In the audit log, an agent appears under its own identity.
Content from your documents goes into the prompts as marked data, kept apart from the instructions. OdySecure validates the output of the derivation and of the agents against a fixed schema. OdySecure checks without AI whether the passage a proposal quotes appears in your document. If it can't be found there, the proposal stays locked until a person decides. In standard operation, the language models run on servers we operate ourselves.
What stays with people
An Agentic ISMS takes work off your team. Duties and responsibility for information security remain with your organisation and its management.
Steering the ISMS stays with people. They set the security objectives, define the criteria for assessing risks and decide which requirements apply to the organisation. In the management review, leadership sets priorities and resources.
Then there are the individual decisions: whether a proposal takes effect, whether a risk is accepted, when an incident is closed and who gets which permissions. In OdySecure, a risk acceptance needs a justification and approval from a second person. A person sets how far each agent may act on its own.
In a vCISO mandate, a senior CISO from Cybervize leads the implementation in OdySecure and keeps your management informed. Responsibility stays with your management even then.
How an auditor sees it
Clause 7.5.2 of ISO 27001 requires documented information to be reviewed and approved. The standard doesn't prescribe a tool. In our reading, it makes no difference to this requirement whether a person or an AI wrote the draft.
In an audit, what counts is whether the measures are in place and work. OdySecure rates how well a measure works from evidence, such as KPIs, results from connected tools like Prowler or Jira, and completed tasks. Measured values count for more than a mere declaration.
For documents, what counts is who reviewed and approved what, and when. OdySecure logs changes with the time, the editor and the old and new value. For AI proposals, it stores the model, the reasoning, their supporting passages in the documents and who reviewed them. The log can only be appended to and is sealed every hour as a hash chain, which a verification package lets you recompute outside OdySecure. Policies go through draft, review and approval under your organisation's approval model. By default, that takes two different people, more than the standard requires.
Where OdySecure stands today
The first two modes, Derive and Ask and instruct, are available in OdySecure. For acting through connected agents, the design-partner programme is under way.
The more complete your documents are, the more OdySecure derives. What the derivation knows about your organisation comes from what you upload and enter. If information is missing, for example on a service provider named in a contract but not in the supplier register, OdySecure creates a clarification task for it.
An Agentic ISMS steers information security and provides the evidence for it. Detecting and fending off attacks is the job of your technical security systems, and their results can feed into OdySecure as evidence.
The comparison with classic ISMS software and the answers to common questions are on our Agentic ISMS page.
This describes OdySecure as measured against the shipped source code on 27 October 2026. Which functions an installation uses depends on its configuration. Standard: ISO/IEC 27001, Clause 7.5.2.