Skip to content
Menu

Praxenor / Veristade / Product explorer

UNDERSTAND THE DIFFERENCE

Follow an action.
Examine the evidence.

An AI tool drafts an internal report, then proposes sending it outside the team. Should that action happen? Separately, what would the activity records tell a reviewer? These six fictional examples show the two decisions. They do not send messages, scan systems or run the products.

VERISTADE / INSPECT THE LAYERS

Inspect the request,
permission and records.

Choose a product, then a layer. Each layer explains one part of the decision. Use the 3D view or switch to a flat view. You can use the controls by touch or keyboard.

Interactive example · Not the working product
Choose the question
Choose the view
Choose a layer to inspect

AGENT DEFENDER / LAYER 01

What is being requested?

Start with what the AI tool wants to do. In this example, it wants to send an internal report outside the team. A reason for sending it is not the same as permission.

See what the product covers ↗

AGENT DEFENDER / LAYER 02

What was the tool allowed to do?

Check what was allowed. Permission to write a draft is not permission to send it outside the team. A tool that does not use the permission check is not covered.

See what the product covers ↗

AGENT DEFENDER / LAYER 03

What should the reviewer see?

Keep a record of what the tool asked to do, what was allowed and what happened. A person can then check the decision and its reason. This example doesn’t give permission or carry out an action.

See what the product covers ↗
Choose a layer to inspect

SWARM HUNTER / LAYER 01

Start with records you’re allowed to use.

Check where the records came from, when events happened and whether any records are missing. These layers only explain the idea. They aren’t a live feed and don’t prove that AI caused an event.

See what the product covers ↗

SWARM HUNTER / LAYER 02

Read the events in context.

Does the activity fit the task? Could there be an ordinary explanation? A change in where information is sent may need a closer look. Several AI tools working together is not, by itself, a sign of misuse.

See what the product covers ↗

SWARM HUNTER / LAYER 03

Separate a finding from a verdict.

Explain what the records show, what is still unknown and what else you need to know. This map helps explain an idea; it does not prove intent or show a physical network. A person must review the findings.

See what the product covers ↗

Reduced motion is enabled: flat, static layers are shown. Layers are conceptual, not a working integration or physical network. No agents, accounts or devices are connected. Nothing is scanned, approved or blocked here.

01 / AGENT DEFENDER

May this
action proceed?

In development

Start with what the AI tool wants to do and what it is allowed to do. Agent Defender only checks tools and actions set up to use it.

Explore Agent Defender Read its technical brief

These examples explain the decisions. They don’t run the product, prove protection or show that it’s ready for general use.

CHOOSE AN EXAMPLE · FICTIONAL RECORDS

A request the tool has permission to make.

The task and the proposed action line up in this example.

Task
Summarize an internal document.
Connection
The proposed action is routed through a connected tool.
Permission
Read the approved document; prepare an internal draft.
Example decision: the requested action is allowed.

The check applies to this action through this tool. It does not cover everything the AI tool could do elsewhere.

The destination changes the decision.

Permission to write a report does not mean it can be sent outside the team.

Task
Prepare an internal report.
Proposed action
Send the report to an external recipient.
Permission
Internal preparation only; external sending is not granted.
Illustrative next step: seek human approval.

Writing a report and sending it are different actions. The tool needs separate permission to send it. This example doesn’t send anything or grant permission.

An action the check never sees.

The tool can only check actions that are sent through it.

Task
Prepare an internal report.
Proposed action
Use a tool that has not been connected for review.
Coverage
No permission decision is demonstrated for this action.
Not covered: the action did not reach the check.

Agent Defender does not control every part of a computer. If an action does not go through its check, it is outside its coverage.

02 / SWARM HUNTER

What do these
events show?

In development · Local review

Start with records you’re allowed to use. Look at how the events relate and where the information came from. Explain the concern and what you still don’t know.

Explore Swarm Hunter Read its technical brief

These fictional examples are not live threat reports or customer results. They do not prove protection or identify who caused an event.

CHOOSE AN EXAMPLE · FICTIONAL RECORDS

Working together can be normal.

Several AI tools working together can be part of an ordinary task. Teamwork alone does not prove misuse.

Record 01
Reader reviews an approved source.
Record 02
Analyst prepares the requested internal summary.
Context
The work and destination match the documented task.
Illustrative finding: no concern established by teamwork alone.

Check permissions and where the records came from. Related activity can have an ordinary explanation, but this example does not prove it is safe.

A sequence worth investigating.

Look at the original record, what happened next and where the information might be sent.

Record 01
Reader opens an approved internal document.
Record 02
Analyst prepares a summary for internal use.
Record 03
Sender proposes an external destination.
Illustrative finding: review the changed destination.

Check the concern against the records, the original task and any separate permission. The sequence does not by itself prove an attack or that AI was involved.

Missing context changes the conclusion.

Two events can look similar without being connected. When information is missing, say what you don’t know rather than fill in the gaps.

Available
Two records reference a similar destination.
Missing
Reliable times, permissions and information about where the records came from.
Next question
What permitted records could confirm or challenge the connection?
Illustrative finding: unresolved.

Keep the unknowns visible and consider other explanations. A person reviews the findings. Swarm Hunter does not automatically block activity or change permissions.

03 / THE OTHER PRODUCT DIRECTIONS

What about the other products?

Malware Scout is experimental, and Stillpoint is paused. You can read about their plans and limits here, but there’s no working product demonstration.

Experimental · In development

Malware Scout

An experimental tool for reviewing a Windows security concern. It isn’t antivirus. Scanning, monitoring and removal still need development and testing.

See what Malware Scout covers
Development paused

Stillpoint

A planned AI workspace for sensitive work on your own computer. Development is paused. It is not an available service, and its privacy has not been independently proven.

Read about the Stillpoint direction

BRING YOUR QUESTION

Which decision does
your team need to make?

Describe the general task and the decision you need to make. Keep passwords, confidential records, customer data and suspicious files out of the contact form.