Skip to content

How it works

One decision, at the moment before commit.

Agents and people act. SentinelX decides whether a protected action is admissible before it commits, and leaves evidence of the decision.

Permission is not authority.

Detection observes failure. Enforcement prevents it.

Prevention holds when SentinelX is placed in the commit path.

Replaces nothing

Your identity, endpoint and detection tools stay.

Fails closed

No decision means no commit.

Evidence of every decision

Allowed, refused or held.

1. Where it sits

After IAM, EDR and detection. Before the action is real.

Your access, endpoint and detection tools stay exactly where they are. SentinelX takes the last position: after authorization, before execution. We put the boundary where the action commits and take write access away from everything else.

Where the boundary sits

Illustrative

A person or an agent

asks to do something that matters

Identity and access
Endpoint
Detection

Your existing tools, unchanged

SentinelX

after authorization, before execution

Where it commits

the ledger, the orchestrator, the gateway

One writer per consequence.Other paths that could write:ScriptsService accountsAdmin consoleslose write access.

2. What we protect

The actions you cannot take back.

Not every action needs a boundary. The ones that move money, destroy data, hand out power, or touch the physical world do.

Money movement

Illustrative

A payment to a new payee, or a change to where invoices get paid.

The request

Change the bank details for a supplier, then pay the next invoice.

One possible outcome

Held

3. Three outcomes

Committed, refused, or held.

Every protected action ends in one of three plain outcomes. Each one leaves evidence of the decision.

Committed

The action is admissible. It goes through as requested, and the decision is recorded.

For the requester: The requester sees nothing new. Work continues.

Refused

The action is inadmissible. It does not happen, and the decision is recorded.

For the requester: The requester is told it was refused and what was refused, never left guessing.

Held

Nothing commits yet. The action waits until what it needs is there, such as a second approval.

For the requester: The requester can supply what is missing. Until then, nothing changes.

4. Fails closed

No decision means no commit.

A timeout, an error, or a missing decision is treated as a refusal. The action does not happen.

Failing closed is a property of where the call is placed. Because the boundary sits in the commit path, there is no way around it when it is quiet. A control that sits beside the path can only watch.

5. When we don't know

What we do when we don't know.

Uncertainty is never read as a yes. In each of these cases the action is treated as inadmissible and it does not commit.

No proof

The request cannot show it has the authority it claims.

Inadmissible. The action does not commit.

Ambiguity

The request could mean more than one thing.

Inadmissible. The action does not commit.

Incomplete context

Something needed to decide is missing.

Inadmissible. The action does not commit.

An unknown change of state

The target changed in a way we cannot account for.

Inadmissible. The action does not commit.

6. What each seat gets

One boundary, four views of it.

01

Board & CEO

A short list of things that must never happen in the business, and a boundary that holds them when placed in the commit path. Policy becomes a line, not a hope.

02

CISO

Control after IAM, EDR and detection: after authorization, before execution, including actions taken by AI agents holding valid credentials. Nothing you run today is replaced.

03

CIO & Chief Architect

You own the write path. No application is rewritten. For each protected action there is one writer, one check, and evidence of the decision.

04

Compliance & Audit

Evidence of every decision, allowed or refused, instead of logs written by the system that let it through.

More on each seat: Who it's for.

7. Getting started

From a short list to a boundary in place.

No new process for your teams. We start with what matters most and widen from there.

    01

    Map the action paths

    Together we list the actions that matter and find where each one actually commits.

    02

    Name what must never happen

    In your words, not code. The list grows as you add more of the business.

    03

    Place the boundary in the commit path

    Each protected action gets a decision before execution. Other paths lose write access.

    04

    Run and review the evidence

    Every decision leaves evidence your team can review, alongside the tools you already use.

8. Questions

Plain answers.

If yours is not here, bring it to the demo.

What happens if SentinelX does not answer in time?

Nothing commits. A missing decision is treated as a refusal. That holds because of where the call is placed: in the commit path, not beside it.

Does it replace IAM, EDR or detection?

No. Those keep doing their jobs. SentinelX sits after them, at the last step before a protected action becomes real.

What about AI agents?

An agent holding valid credentials is treated like anyone else. Its protected actions get the same decision before they commit.

Do we have to rewrite our applications?

No. The boundary goes where the action already commits. Your teams keep their systems.

Does it read prompts or conversations?

No. It looks at the action, not the conversation behind it.

Can it undo something that already happened?

No. It decides before the action commits. That is the point: some actions cannot be taken back.

See it on one of your own actions.

Bring the action you worry about most. We will walk it to the commit point with you.

Book a demo