AI security assessment with a controlled path to protection.

Understand the environment before changing it

The first job is to establish what exists and how it connects. The specified assessment covers hosts, networks, applications, identities, databases, cloud workloads and the systems that build and deploy software. An isolated package warning is useful evidence; knowing where that package runs, what reaches it and which privileges it holds makes the evidence actionable.

Discovery is bounded by the customer’s authorised scope and available access. Assets that cannot be inspected must remain recorded as coverage gaps. An unreachable system cannot be marked secure merely because a collector returned no findings.

• Read-only first-entry assessment across 25 defined assessment and protection areas.
• Contextual risk correlation using exposure, privilege, exploit evidence and business importance.
• A client report that separates evidence, interpretation, recommendations and uncertainty.

An explicit boundary between advice and action

AI Defence is designed to explain a recommendation before requesting permission to apply it. Customers can approve selected findings, individual changes, a clearly defined bundle, or no changes. Approval should identify the target, action and permitted scope; it is not a standing instruction to execute arbitrary commands.

Material changes require a current-state check, dependency analysis, an assessment of possible impact, a rollback plan and a way to preserve administrator connectivity. A proposed firewall improvement that removes the only management path is a failed change even if the intended port closes.

Evidence continues after a remediation attempt

The intended workflow retests the security finding and relevant service health after a change. A patch installation, successful command or AI-generated explanation does not by itself establish that a vulnerability was resolved. Failed verification keeps the issue open for investigation.

An accepted baseline records expected assets, users, ports, packages, workloads, configuration and approved exceptions. Continuous monitoring then looks for new vulnerabilities and changes against that baseline. Suspicious observations create evidence for investigation; they do not prove compromise on their own.

The relationship with Remedy and Nexus

Remedy’s software debugger addresses defects through diagnosis, source-code repair and verification. AI Defence adds a distinct security assessment and protection scope. Where a finding requires an application repair, the two workflows can be designed to meet through a governed hand-off.

Nexus is Altari’s broader autonomous software engineering control plane. These product roles are complementary. This website describes AI Defence’s September 2026 specification; it does not claim that every collector, integration or remediation action is generally available.