How Encased lines up with the Australian Signals Directorate's Guidelines for secure AI system development. Self-assessed 11 November 2026.
Alignment self-assessment by the Encased team — not an ASD endorsement, certification or independent review. The guidelines are voluntary guidance; nobody is "certified" against them. Read the official source at cyber.gov.au.
1. Secure design
Understand risks and threats before building, and design for security from the start.
Model threats to the AI system and its usersPublished threat model for agent actions: injection, exfiltration, runaway loops, destructive actions. · Containment referenceIn place Limit what the AI system can do (least privilege)Policies check submitted proposals; configured review-required actions are held. The integration must enforce the decision. · Runtime guardrails guideIn place Consider security trade-offs of the designIsolation controller keeps credentials and effects outside the agent; limits are documented openly. · Known limitationsPartial 2. Secure development
Secure the supply chain, document assets and manage technical debt.
Secure the software supply chainIsolation package requires reviewed, digest-pinned container images; packages are hashed. · Builder assessmentPartial Document data, models and promptsPublic REST contract and deployment reference; formal asset register not yet complete.Partial
Test before releaseLinux Docker probes passed 9/9 on 10 November 2026 (Ubuntu 24.04.5 · Docker 29.1.3 · cgroup v2), with 0 failed and 0 skipped. This is builder evidence, not independent review or production qualification. Independent review and deployment-specific qualification remain pending; this server run does not establish on-device isolation. · MilestonesPending review 3. Secure deployment
Protect infrastructure and models, plan for incidents, release responsibly.
Secure infrastructure and accessEncrypted traffic, hashed agent keys, database-level workspace separation. · Security roadmapIn place Fail safelySDKs block actions by default if Encased cannot be reached.In place
Incident managementNotifiable Data Breaches process stated; formal incident response plan being finalised. · Compliance statementPartial Independent assurance before releaseNo independent review yet. Builder assessment says not accepted for production containment. · Review checklistPending review 4. Secure operation and maintenance
Monitor behaviour, log inputs and actions, update securely and share lessons.
Monitor system behaviourDashboard of recorded workspace proposals and decisions, with human review. Key revocation blocks future gateway requests, not running processes. · DashboardIn place Keep logs for investigationContainment and policy changes recorded in an audit log that cannot be edited.In place
Vulnerability disclosurePublic reporting address with reasonable fix time requested. · Report a vulnerabilityIn place Other Australian obligations
Privacy Act 1988 and Australian Privacy PrinciplesIn place (basic)Collect only what is needed; no sale of personal information; full privacy policy being finalised.
Notifiable Data Breaches schemeIn placeAffected customers and the OAIC notified where required.
ASD Essential EightPlannedMaturity review of Encased's own systems planned; not yet assessed.
Voluntary AI Safety Standard (DISR)PlannedMapping of Encased's guardrails to the standard's guardrails planned.
Need this for a security questionnaire or tender? Email get@encased.ai or book an AI agent security audit.