AI Defence / Remedy architecture
Make trust specific, explainable and revocable.
These principles guide the Remedy Defence architecture and the questions used to evaluate it. They are design requirements, not a substitute for implementation evidence or a guarantee against every attack.
Measure the state and grant the minimum authority
Trust applies to a defined user, object, device, service or action under current conditions. Names, labels and network membership are insufficient. The intended decision is supported by measured state, provenance and policy.
Least privilege means the task receives only the capability it needs. A document does not need broad network authority to be read. An administrator’s management session does not need permanent unrestricted access to every service.
Contain uncertainty and withdraw inherited trust
Unknown content may be needed for legitimate business. It should have a path for safe evaluation without inheriting unrestricted rights. Isolation and bounded analysis protect the surrounding environment while uncertainty remains.
When material unexplained change or new risk invalidates earlier evidence, the old decision should not persist automatically. The resulting state can be pending, restricted or quarantined until the required revalidation supports a new decision.
Check identity, posture and the action independently
Authentication establishes something about a user or session. It does not prove that the device is healthy or that the requested operation should be allowed. Successful MFA must not cancel an independent posture failure.
Public services are deliberately exposed for their intended audience. Private services default to denial except through approved access paths. A known address alone cannot justify private access.
Separate production and deception completely
Deception can supply useful observations only when its synthetic resources have no production data, useful credentials, management authority or path back into the real estate. Production denial must remain effective if the decoy is unavailable.
The intended scope is defensive observation. No response should depend on attacking an external system or attributing a human identity solely from an address.
Keep AI advisory and actions governed
Deterministic controls should act on defined conditions such as an invalid signature, prohibited path, revoked credential or known malicious hash. AI can help interpret ambiguous evidence and explain a recommendation. It should not become an unbounded authority to execute commands.
Automation belongs within supported capabilities and tenant policy. Higher-impact work, restoration, rebuilding and major production changes may need explicit approval. The record should identify the action, target, authority and result.
Build recovery and verification into the lifecycle
Assume some attacks will succeed somewhere. Protected backups, recovery access and tested procedures need planning before an incident. Reconnecting an endpoint is not the same as restoring its data or resolving the cause of compromise.
A remediation is not complete merely because an action succeeded. The original condition and relevant service health need fresh assessment. Missing, conflicting and inconclusive evidence should remain visible rather than being converted into a reassuring closure label.
Questions answered
Talk directly to Altari
Bring the environment.
Start with the right questions.
Discuss a demonstration, current capabilities and a bounded evaluation with Altari Systems.