Security and governance
Autonomy is only credible with the controls published.
Your working capital system holds the most sensitive financial data in the business, and now it acts inside it. Two things must hold: the data stays protected, and every autonomous action is bounded, logged, and reversible.
AI governance
How does Kapittx govern what AI agents are allowed to do?
Every agent capability is assigned an autonomy level (assist, automate, adapt, act, or escalate) and operates inside an envelope the customer defines by value threshold, segment, and permitted action type. Actions below a confidence threshold escalate to a person instead of executing.
Each action is recorded with the data that informed it, the envelope it acted under, and the resulting ERP posting. In practice this produces a more traceable record than the spreadsheet-and-shared-inbox process it replaces.
Certifications
Independently audited, continuously verified.
- 01
SOC 1 Type 2
Controls over financial reporting, the ones your auditors will ask about when a system touches the ledger.
- 02
SOC 2 Type 2
Audited controls over security, availability, processing integrity, confidentiality, and privacy of customer data.
- 03
ISO 27001
An information security management system established, implemented, maintained, and continually improved across operations.
Certifications
The marks behind the claims.
Security in depth
Defence at every layer.
- 01
TLS 1.2 in transit
All traffic between you and Kapittx runs over Transport Layer Security 1.2.
- 02
AES-256 at rest
Data is encrypted at rest using a 256-bit standard.
- 03
Backed up, encrypted
Continuous backups carry the same encryption guarantees as the primary system.
- 04
Continuous monitoring
A dedicated security team watches for emerging threats and updates controls in response.
Agent controls
What bounds an autonomous action.
These are configuration, not policy statements, each one is something your team sets and can change.
- 01
Autonomy level per capability
Each agent capability is explicitly assist, automate, adapt, act, or escalate. Nothing is autonomous by default.
- 02
Approval envelopes
Value thresholds, customer and vendor segments, and permitted action types define where an agent may act unattended.
- 03
Confidence thresholds
Below the threshold, an agent routes to a person rather than acting on a weak inference.
- 04
Full action audit trail
Every action records its inputs, the envelope it acted under, and the resulting posting.
- 05
Reversibility
Agent actions can be reversed, and overrides are captured alongside the original recommendation.
- 06
Segregation of duties
Approval routing respects your existing delegation of authority; agents do not collapse it.
FAQs
Questions procurement will ask.
- Is our data used to train models shared with other customers?
No. Customer data is not used to train models shared across customers.
- Can we get the SOC reports and a completed security questionnaire?
Yes, contact us and we will share the SOC reports, our SIG, and answer your security team's questions directly.
- Where is our data hosted?
Hosting region is confirmed during procurement, as it depends on your entity structure and regulatory requirements.
- What access does Kapittx staff have to our data?
Access is role-based and logged, and production access is restricted to what is required to operate and support the service.
- How do we prove to an auditor what an agent did?
Each action carries its full audit trail (inputs observed, decision taken, approval envelope, and resulting ERP posting) exportable for audit.
- Who can change an agent's approval envelope?
It is permission-controlled and every change is logged with the person and timestamp, so widening what an agent may do unattended is itself an auditable event rather than a silent configuration tweak.
Need our security pack for procurement?
Get in touch, we'll share SOC reports, our SIG, and walk your team through the agent controls.








