Security & Compliance

Built for CFOs who get audited.

An AI agent writing to your SAP system is a serious decision. Here is exactly how we handle confidence, control, audit trail, and credential security — no marketing language.

01

Confidence scoring

Every invoice gets a score before it touches SAP

Capped AI computes a composite confidence score across five extraction components: document quality, vendor name match, field completeness, tax code validity, and amount consistency. Only invoices above your configured threshold are auto-posted.

  • Composite score = weighted average of 5 independent components
  • Configurable posting threshold (default: 0.85) — you set the bar
  • Sub-threshold invoices routed to exception queue, not discarded
  • Score breakdown shown per-invoice in the audit trail
02

Shadow mode

We run in parallel for 30 days before writing a single SAP document

During the pilot, Capped AI processes invoices alongside your existing AP team. It predicts what it would post — but does not post. You compare outcomes. When accuracy is validated on your real data, you flip the switch.

  • Full extraction + confidence scoring runs silently in parallel
  • Daily accuracy report vs. what your clerks actually posted
  • No SAP writes until you explicitly authorize production mode
  • Shadow period can be extended indefinitely — your call
03

Never autonomous on sensitive postings

Hard rules enforced at the infrastructure layer

Certain invoice categories require human approval regardless of confidence score — these are infrastructure-layer rules, not AI judgment calls.

  • New vendor (not in SAP vendor master) — always human approval
  • Invoice amount > $10K — human approval required
  • Tax-sensitive postings — human approval
  • Credit memos and debit notes — human approval
04

Audit trail

Every Fiori action logged for 7 years — SOX-grade, immutable

The Capped AI audit log is append-only. Every field extracted, every confidence score computed, every Fiori action taken, every screenshot captured, every human approval or rejection — timestamped, actor-attributed, and cryptographically signed at write time.

  • Per-step screenshot of every SAP Fiori action during posting
  • 7-year retention policy (configurable up, not down)
  • Append-only: records cannot be altered or deleted
  • SAP document number cross-referenced on every posting entry
  • Exportable to CSV for external audit or SIEM integration
05

Credential security

AES-256-GCM encrypted SAP credentials

The SAP service account credentials used by the browser agent are encrypted at rest with AES-256-GCM. No SAP data leaves Capped AI infrastructure beyond what is necessary to complete the posting.

  • AES-256-GCM encryption for stored credentials
  • Credentials never logged or stored in plaintext
  • Separate encryption key per customer tenant
  • Access is scoped to the minimum required SAP permissions
06

Compliance

Compliance certifications planned post-launch

We are a design-partner stage company. Our security practices are built to enterprise standards. Formal compliance certifications are planned for after the V1 launch.

  • Internal quarterly security reviews against Trust Services Criteria
  • Penetration test report (Q1 2026) available to enterprise customers under NDA
  • SBOM and dependency manifest available on request
  • Compliance certifications planned post-launch

Available on request

Penetration test report (last conducted Q1 2026), Software Bill of Materials (SBOM), dependency manifest, vulnerability disclosure policy, and data processing agreement (DPA) — available to qualified enterprise customers under NDA.

Speak to our security team

If your IT, InfoSec, or internal audit team needs a deeper briefing, architecture diagram, or compliance documentation, reach out and we will schedule a technical call.

Request a security briefing

Or email us directly at security@capped.ai