Security
Built so you can trust it with real work.
AI that can act needs tighter controls than software that only displays. These are the controls we put in place before a workflow touches your data, and the ones that protect this website.
AI-specific defences
Prompt injection defence
Emails, PDFs and web pages are treated as untrusted data. Content can never change instructions, call tools or widen permissions.
Rules decide, models read
Models extract and draft. Explicit, inspectable rules make the decisions, so every outcome can be explained.
Output validation
Every model output is checked against a strict schema before anything downstream acts on it.
Spend and volume limits
Token budgets, sending limits and transaction caps are enforced outside the model, so a runaway loop stops itself.
Access and secrets
Least privilege
Each workflow gets its own credentials with the narrowest scope it needs. No shared admin keys.
Secrets vault
Credentials live in a managed vault, are never written to code or logs, and are rotated on a schedule and on staff change.
Separate environments
Development, staging and production are isolated. Test data never touches production systems.
Strong authentication
Multi-factor authentication on every account that can reach client systems.
Data protection
Redaction before models
Sensitive fields such as account numbers and health identifiers are masked before a document reaches a model.
Never used for training
Client data is not used to train models. We select providers whose terms say the same.
Isolation per client
Each client has separate storage, keys and workflow runtime.
Encryption and retention
Encrypted in transit and at rest, with retention windows you set and deletion on request.
Oversight and recovery
Human approval gates
Payments, contracts, adverse decisions and anything irreversible wait for a named person.
Full audit trail
Source, rule, model output and approver are recorded together and exportable for your auditors.
Kill switch
Pause any workflow in one click, with queued work held safely until you resume.
Shadow mode first
New workflows run beside your team and report accuracy before they are allowed to act.
This website practises what we sell.
Content Security Policy
A strict, per-request nonce policy blocks injected scripts, inline handlers, framing and unexpected connections.
Transport security
HTTPS only, with HSTS preload and a no-sniff policy.
Isolation headers
Frame denial, cross-origin opener and resource policies, locked-down browser permissions and a minimal referrer policy.
Hardened booking form
Same-origin checks, JSON only, size limits, strict validation, honeypot and timing checks, optional bot verification and per-address rate limits.
No sensitive logging
Booking details are never written to logs. Rate-limit keys are salted hashes, not addresses.
Safe email
All user input is escaped in messages, headers are stripped of line breaks and calendar files are escaped to the standard.
Dependency checks
Dependencies are pinned and audited on every release.
Disclosure policy
A published security contact and a safe-harbour promise for good-faith reports.
Compliance, plainly.
We design to the standards your industry already holds you to, and we do not claim certifications we have not earned.
| Privacy | Designed around PIPEDA and, where relevant, GDPR principles: purpose limitation, minimisation and deletion on request. |
|---|---|
| Healthcare | Workflows handling health information are scoped with your privacy officer, with PHIPA and HIPAA requirements built into the design. |
| Financial | Controls mapped to your audit needs, including segregation of duties, approval limits and immutable records. |
| Assurance | We follow practices aligned with SOC 2 controls. We will share our current status and questionnaire answers on request. |
| Data location | Canadian data residency can be arranged where your requirements call for it. |
Found something?
We welcome responsible disclosure. Report a vulnerability in good faith and we will acknowledge it, fix it and not pursue action against you.
- Email admin@baltar.ca with the details
- Our contact file is at /.well-known/security.txt
- Please do not access other people’s data or disrupt the service