Supplier security summary
A short factual summary for IT and procurement teams assessing AIK as a supplier, including organisations in scope of NIS2. Every item below describes behaviour that is implemented in the product today.
The single most important control, and the one most suppliers cannot offer.
AIK runs entirely inside your own network — application, database and documents. Your business data never leaves your building, and the platform works with local AI models, so prompts and documents need not be sent to any external AI provider.
Each user's files sit in a private sandbox directory. File access by the AI agents is fenced to that sandbox; agents cannot read configuration files, keys or other users' documents.
Passwords are stored as bcrypt hashes, never in readable form. Sessions use signed JWT tokens that can be invalidated centrally on logout.
Standard TOTP second factor (Google Authenticator, 1Password and similar), enabled per user.
Login, registration and password-reset endpoints are rate-limited per IP address and per account.
Administrator and standard-user roles. Sensitive operations — system settings, model management, platform updates — are restricted to administrators.
API keys, mail passwords and integration secrets are encrypted before they are written to the database. A stolen database file does not yield working credentials without the separate encryption key.
Server installs terminate TLS at nginx with Let's Encrypt certificates, configured during setup and renewed automatically.
Designed so that a record cannot be quietly altered after the fact — including by us.
Every logged action is chained to the one before it with a SHA-256 hash. Editing or deleting any past entry breaks the chain and is detected.
An administrator can run the check at any time from the platform and see exactly which entry, if any, fails.
Any automated run can produce a signed certificate showing what ran, when, and with what result. It is verified using a public key, so a recipient can confirm it independently.
Documents and emails produced by the platform are marked machine-readably as AI-generated and carry a signature that detects later alteration. People are also told they are dealing with an AI: phone calls open with a spoken disclosure, messages carry a marker, and the embeddable website chat states it in the conversation itself. This is enforced in code rather than in the agent's instructions, so it cannot be edited away. Together these cover both duties in EU AI Act Article 50, in force since 2 August 2026.
Automatic nightly backup, plus a separate one-click undo for platform updates.
Runs every night at 03:00 as a scheduled system service. Each run captures the full database, the configuration file, the signing key and all user files into one dated archive.
The last 14 archives are kept and older ones removed automatically.
Each archive can be copied automatically to storage outside the server — your own cloud storage or another machine — so that losing the server does not lose the backups. This is a one-time setup step and is strongly recommended.
A single restore command rebuilds a working installation from any archive: database, configuration, keys and files.
Before any platform update the system snapshots both the application and the database. If an update misbehaves, one action restores the previous version — code and data.
Each change runs through an automated pipeline before it can ship: code linting, a suite of nearly 800 automated tests, static security analysis, a scan of all third-party libraries for known vulnerabilities, and a database-migration check that verifies an existing installation can be upgraded and rolled back before the change is allowed to ship.
On-premise updates are delivered as packages signed with an Ed25519 key and verified by checksum and signature before installation, so a tampered update is refused.
A user deletes their own account from Settings, and everything goes with it: agents, workflows, settings and stored keys, documents, payment rows and their whole file folder. Run history and the audit log are kept with the person unlinked — that log is a hash chain, so removing a row would break the proof for every entry after it.
Any user downloads everything held about them as one zip from Settings → Security — rows and files, no request to anyone. Stored secrets are excluded on purpose: they remain readable in Settings, where a copy cannot be mislaid.
Conversation history can be set to delete automatically after a chosen number of days, so data is not kept longer than the purpose requires.
Everything above describes the software. When AIK runs on your server we never receive your data at all. When we host the install for you, we are your data processor under GDPR Article 28, and the Data Processing Agreement states what is processed, where it is stored (EU), which sub-processors exist, how a breach is reported — and what is not promised, including that a restore is whole-platform rather than per-client.
ISO 27001 certifies an organisation's way of working, not a piece of software — so no product, including this one, can be "ISO 27001 certified". What a supplier can do is cover controls on your behalf. If you hold the certificate, or are working towards it, these are the Annex A (2022) controls AIK contributes to. Your auditor will still want your own policies, risk assessment and records.
| Annex A control | What AIK provides |
|---|---|
| A.5.15 / A.8.3 Access control | Role separation, per-user file sandbox, and agents restricted to the folders you allow |
| A.5.16 / A.5.18 Identity & access rights | Named accounts, administrator role separation, sessions that expire, and instant revocation of all sessions |
| A.8.5 Secure authentication | Hashed passwords, TOTP two-factor authentication, and rate-limited login |
| A.5.33 Protection of records | The activity log is SHA-256 hash-chained, so an altered or deleted entry is detectable |
| A.5.28 Collection of evidence | Any run can be exported as a signed Action Certificate; the chain can be verified on demand |
| A.8.15 Logging | Every action recorded with actor, time and outcome, in an append-only log |
| A.8.16 Monitoring activities | Run history, failure alerts, and a health check across every tool server |
| A.8.13 Information backup | Nightly database and file backup, retention limits applied locally and off-site, restore in one command |
| A.8.24 Use of cryptography | Credentials AES-256 encrypted at rest, TLS in transit, Ed25519 signatures on updates and receipts |
| A.8.8 Technical vulnerabilities | Every third-party library scanned for known vulnerabilities on each change, before it can ship |
| A.8.25 / A.8.28 Secure development | Linting, static security analysis and an automated test suite gate every change |
| A.8.29 Security testing | Nearly 800 automated tests, plus a migration check proving an existing database upgrades and rolls back |
| A.8.31 Separation of environments | Development, test and production are separate installations with separate databases |
| A.8.32 Change management | Signed update packages, a snapshot before every update, and rollback covering code and database |
| A.8.9 Configuration management | Settings held in the database and versioned with the install; no per-machine hand editing |
NIS2 Article 21 lists the baseline measures in-scope organisations must have. Where AIK forms part of your supply chain, this is what the platform contributes.
| Measure | What AIK provides |
|---|---|
| Incident handling | Tamper-evident activity log with independent verification; per-run records showing what happened and when |
| Business continuity & backup | Automatic nightly backup with off-site copy, one-command restore, and update rollback covering code and database |
| Supply chain security | On-premise option removes the vendor from the data path entirely; third-party libraries scanned for known vulnerabilities on every change |
| Secure development & disclosure | Automated linting, testing and static security analysis before release; updates cryptographically signed |
| Access control & authentication | Hashed passwords, TOTP two-factor authentication, rate-limited login, administrator role separation |
| Cryptography | Stored credentials encrypted at rest; TLS in transit; signed audit receipts and update packages |
| Effectiveness assessment | The audit-chain check and the automated test suite can be run on demand as evidence |