Security & Compliance

Built securely.
Verified daily.

The platform is aligned to NIST Cybersecurity Framework 2.0 and the ASD Essential Eight. Autonomous scanning runs every night. Every credential is vaulted. Every AI output carries a forensic audit trail.

0
Open Alerts
14
Daily Scan Checks
03:00
Scan Time (AEST)
22 April 2026
Last Full Audit
2027
ISO 27001 Target

Built on recognised standards

We align to internationally recognised security frameworks — not bespoke checklists. This gives clients a common language for assessing posture and comparing providers.

Current — International
NIST CSF 2.0
The US National Institute of Standards and Technology Cybersecurity Framework 2.0 (published February 2024). The global benchmark for organisational cybersecurity posture — adopted by government, finance, healthcare, and technology sectors worldwide. Explicitly covers AI security (OWASP LLM Top 10 mapping added April 2026).
Govern Identify Protect Detect Respond Recover
Current — Australian
ASD Essential Eight
The Australian Signals Directorate's eight mitigation strategies. Mandatory for Australian government agencies; broadly adopted across Australian private sector as a baseline security benchmark. High credibility with Australian financial services, legal, and government clients. Aligned at Maturity Level 1–2.
Application Control Patch Applications Restrict Admin MFA Backups
Roadmap — 2027
ISO/IEC 27001:2022
The international gold standard for Information Security Management Systems (ISMS). Requires formal audit and certification by an accredited body. Targeted when the platform reaches sustained government or ASX-listed client revenue — at which point the certification cost is justified by contract win rate and procurement requirements.
Formal Certification Accredited Audit ISMS

Six layers of protection

Security is not a checklist item — it is built into every layer of the platform architecture, from credential storage to AI output governance.

🔑
Credential Management
All credentials stored in a permission-enforced vault (chmod 600, never in code or version control). Verified absent from git history on every scan. Secrets are passed via environment variables only — never as process arguments (which would expose them in system process lists). Rotation schedule enforced per credential class. Autonomous scan auto-corrects permission drift.
Vault-based Git-clean Auto-correcting
🌐
Network Security
Minimal external attack surface. All public-facing services route through Cloudflare Tunnel — no inbound ports exposed directly to the internet. Every API service requires authentication. SSH is key-only (password authentication disabled). Each world-open port is individually justified, documented, and verified on every scan.
Cloudflare Tunnel Key-only SSH Per-port auth
🤖
Autonomous Monitoring
A 14-point security scanner runs automatically at 03:00 AEST every night. Checks include: credential file permissions, git secret exposure, credential patterns in logs, unexpected open ports, rogue services, default passwords, SSH hardening, CORS configuration, and JWT secret persistence. Auto-remediates where safe. Alerts immediately on anomalies. Silent if everything is clean.
Nightly Auto-remediate 14 checks
🏠
Local-First AI
Sensitive research, client materials, political strategy content, and medico-legal data are processed on-premise using locally hosted LLMs (Ollama). No sensitive data is transmitted to third-party model providers. Cloud models are used only for non-sensitive reasoning tasks where no client-identifiable content is involved.
On-premise LLMs No data egress Client-isolated
📋
Governance & Audit Trail
Every AI-generated recommendation passes a mandatory governance pre-check across four risk domains: reputational, legal, financial, and political. BLOCK outputs are never delivered and are forensically logged. FLAG outputs require adviser sign-off before delivery. CLEAR outputs carry a full governance attestation. Adviser compliance rates are tracked over time. Complete audit trail on all AI outputs — no exceptions.
BLOCK/FLAG/CLEAR Forensic log Adviser scoring
🚨
Incident Response
Documented runbook for three incident classes: credential compromise, unauthorised network access, and rogue automated agents. Each includes: immediate containment, 30-day log audit, affected-party notification, and mandatory protocol update. Post-incident reviews are required — not optional. Response procedures are tested as part of infrastructure changes.
Documented runbook 30-day audit Post-incident review

14 checks, every night

The security scanner runs as an isolated agent at 03:00 AEST. It auto-fixes what it can, kills rogue services it detects, and sends an alert only if it finds something that requires attention. If everything is clean, it logs silently and stops.

Last scan: 23 April 2026 — clean
Security scan completed 23 April 2026 at 03:00 AEST. All 14 checks passed. Autonomous nightly scanner active — no alerts.
0 Open Alerts Daily Active 03:00 AEST

Findings & remediation log

All security findings are documented with severity, status, and remediation detail. Nothing is marked resolved without verification.

Date Finding Severity Status Remediation
22 Apr 2026 Android Control Service — no authentication on 20 endpoints including shell execution and device control. World-open on port 8765. Critical ✓ Fixed X-API-Key header auth added at app level. Key stored in secret vault.
22 Apr 2026 /tmp HTTP server running since Apr 20 — serving entire /tmp directory (incl. OAuth codes) world-open on port 8877. Critical ✓ Fixed Process killed. Scanner now detects and kills rogue servers automatically.
22 Apr 2026 Cloudflare tunnel token exposed as CLI argument — visible in process list to any local process. Critical ✓ Fixed Systemd unit rewritten to load token from EnvironmentFile. Token no longer in process list.
22 Apr 2026 Four secret files with 664 permissions (world-readable). High ✓ Fixed All enforced to 600. Scanner auto-corrects permission drift on every run.
22 Apr 2026 Pollster Portal CORS configured with wildcard origin and credentials=True. High ✓ Fixed Replaced with explicit allowed origins list.
22 Apr 2026 JWT secret ephemeral — regenerated on every process restart, invalidating all sessions. High ✓ Fixed JWT secret persisted to secret vault. Loaded at startup consistently.
22 Apr 2026 Prometheus and observability metrics exposed world-open without authentication. Medium Accepted Acceptable within Tailscale network boundary. Flagged for hardening if external access ever enabled.

Pre-defined playbooks

Three documented response playbooks. Each is designed to be executed immediately without waiting for judgement calls under pressure.

🔑 Credential Compromise
  1. Immediately revoke the compromised credential at the provider
  2. Generate new credential and store in secret vault
  3. Restart all affected services
  4. Audit access logs for unauthorised use (last 30 days)
  5. Update rotation schedule and document in incident log
🌐 Unauthorised Network Access
  1. Kill the exposed service immediately
  2. Check access logs for data exfiltration indicators
  3. Rotate any credentials the service had access to
  4. Document in daily memory file
  5. Update scanner to detect the exposure class in future
🤖 Rogue Automated Agent
  1. Disable the cron job immediately
  2. Review its recent run logs for unauthorised actions
  3. Audit outbound messages sent during rogue period
  4. If data was exfiltrated, notify affected parties
  5. Conduct post-incident protocol review — mandatory

Rotation schedule

All credentials are on a documented rotation schedule. High-value credentials rotate more frequently.

Twitter OAuth Tokens
Social campaign accounts. Rotate on suspicious activity or 90-day cycle.
Next due: 22 July 2026
Cloudflare API Token
Full domain and DNS control. 180-day rotation or immediately on suspected compromise.
Next due: 22 October 2026
Android Control API Key
Full device control. 90-day rotation, stored in secret vault.
Next due: 22 July 2026
Pollster Portal JWT Secret
Session authentication for polling portal. 180-day rotation.
Next due: 22 October 2026
Google OAuth Tokens
Calendar, Drive, Gmail, Search Console. Auto-refresh by library — no manual rotation required.
Auto-managed
Anthropic / OpenAI Keys
LLM API access. Rotate on billing anomaly or 180-day cycle.
Next due: 22 October 2026

Steganographic fingerprinting
for leak attribution

Every exported briefing carries an invisible mark — not a watermark you can see and remove, but a fingerprint woven into the encoding of the document itself. If it leaks, we know exactly who had it.

How it works

Every exported briefing is encoded with a unique recipient identifier embedded in Unicode variation selectors, zero-width characters, and subtle whitespace patterns distributed throughout the document. The encoding is invisible in rendering but survives virtually all document transformations — including printing, scanning, and OCR re-digitisation.

Each encoding maps to a specific delivery record in the attribution ledger: the recipient, the timestamp, the scenario reference, the governance status at time of export, and the document version. When a leaked document is recovered — from a screenshot, a forwarded email, a scanned print, or a published excerpt — the platform extracts the fingerprint and identifies the source within seconds.

The 2M+ unique ID space means every individual delivery across the platform's entire client base carries a distinct fingerprint — no recycled IDs, no collisions.

Technical note

Uses Unicode homoglyphs, variation selectors, and systematic zero-width joiner placement — indistinguishable in any rendered view. Survives email reformatting, PDF extraction, and OCR quantisation. The encoding scheme is proprietary and the mapping ledger is stored locally, not in any cloud service.

When to use it

Internal strategy briefings

Campaign-level strategy documents distributed to senior staff, advisers, or principal offices. Any situation where a briefing circulates among more than one person and its contents could damage a campaign if published.

Opposition research

Research products that would be damaging if leaked to the subject. Knowing a copy reached a specific adviser allows rapid containment before exposure.

Coalition documents

Negotiation positions, offer frameworks, and internal assessments shared with coalition partners. The Canary Protocol allows post-hoc attribution even when physical copies are involved.

2M+ unique fingerprint IDs Print & scan survivable Local ledger — no cloud