SIFTR

Security & Data Protection

Enterprise-grade protection for candidate data and hiring workflows.

Security Principles We Follow

SIFTR's security is built around four core principles:

Zero Trust by Default

Every request is authenticated. Every resource is least-privilege accessible. Nothing is implicitly trusted.

Defence in Depth

Multiple layers of security: network, encryption, access control, AI-layer protection, and audit logging.

Privacy-First Architecture

We minimise data early:

  • • PII removal at ingestion
  • • Optional "no raw payload storage"
  • • Configurable retention windows

Compliance by Design

Aligned to:

  • • UK Data Protection Act
  • • GDPR
  • • EU AI Act (High-Risk Systems)
  • • NIST AI Risk Management Framework
  • • ISO/IEC 27001 & 42001 (directionally aligned)

Data Flow Overview

1

Ingestion

When a CV is uploaded:

  • • File is scanned
  • • Parsed into text
  • • PII removed automatically
  • • Only sanitized text enters the AI pipeline
2

Processing

Two-round ethical pipeline:

  1. 1. Round 1: Remove bias proxies
  2. 2. Round 2: Evaluate skills only

Both rounds run inside a locked-down secure environment.

3

Output & Storage

  • • Sanitized content stored separately from raw CVs
  • • Clean JSON stored with strict schema
  • • Audit logs stored in immutable format
  • • Access controlled by recruiter roles
4

Retention & Deletion

Configurable per-agency:

  • • 14 / 30 / 60 / 90 / 180 day options
  • • Optional instant-delete for raw CVs
  • • Secure deletion triggers followed by verification logs

Data Protection Controls

Encryption & Key Management

In Transit

  • • TLS 1.2+
  • • HSTS enforced
  • • Secure ciphers only

At Rest

  • • AES-256 encryption
  • • Separate keys for CV storage vs audit storage
  • • KMS-managed automated rotation

Access Controls (RBAC)

Recruiters can only:

  • • View records they own
  • • Access sanitized evaluations
  • • Add human feedback
  • • Never access audit logs unless permissioned

Admins can:

  • • Configure retention policies
  • • Access audit evidence
  • • Manage users and permissions

SIFTR personnel access:

  • • Restricted by need-to-know
  • • Logged
  • • Covered by contractual AI confidentiality terms

Infrastructure Security

  • AWS-hosted (EU/UK region)
  • Isolated VPCs
  • Private subnets for processing
  • No inbound public connections to model pipeline
  • WAF protection with automated blocking
  • Automated patching for critical CVEs
  • Hardened server images

Data Minimisation (Core to EU AI Act)

We store:

  • • Sanitized CV text
  • • AI evaluation JSON
  • • Audit metadata
  • • Optional recruiter feedback

We do not store:

  • • Gender
  • • Age
  • • Ethnicity
  • • Nationality
  • • Political information
  • • Religion
  • • Family status
  • • Visa info
  • • Any raw AI model embeddings (optional setting)

Optional "No Raw Payload Storage" Mode

Many enterprise customers demand this.

If enabled:

  • Raw CVs are purged instantly after PII stripping
  • Only sanitized text remains
  • Audit logs store checksums, not content

Perfect for banks, public sector, critical industries.

Audit & Monitoring

Immutable Audit Logging

Every run logs:

  • • Timestamp
  • • Schema version
  • • Prompt version
  • • Ethical ruleset version
  • • Bias tokens removed
  • • PII detected
  • • Confidence score
  • • Recruiter feedback (if applicable)

Audit logs cannot be edited or deleted.

Continuous Monitoring

Alerts for:

  • • Suspicious access patterns
  • • Brute-force attempts
  • • Excessive failed logins
  • • Output anomalies
  • • Drift detection flags

All captured via:

  • • CloudWatch
  • • GuardDuty
  • • Automated alerting rules

Incident Response

If an incident is detected:

  1. 1. Containment immediately
  2. 2. Root-cause analysis
  3. 3. Remediation
  4. 4. Report to customer (if required under GDPR)
  5. 5. Full audit trail attached

SIFTR will publish a Security Incident Policy in Q1 2026.

Model Security (AI-Specific Protection)

SIFTR prevents:

  • Prompt injection
  • Output corruption
  • Hallucination drift
  • Unauthorized output modification

Controls include:

  • Strict JSON schemas
  • Sanitized input layers
  • Two-round control pipeline
  • Ruleset versioning
  • Output validation filters

This prevents the majority of modern AI vulnerabilities.

Customer Guarantees

We guarantee:

✔ Data stays in the UK/EU

No processing in US regions unless contractually approved.

✔ No model training on your data

SIFTR never trains models using candidate CVs.

✔ Data ownership stays with the customer

We process it — you own it.

✔ You choose your retention window

Full configurability for compliance needs.

✔ You can export or delete all data

At any time.

Coming Soon

SOC-2 Lite Pack
Vendor Security Due Diligence Pack (VSDD)
Penetration Test Summary
Model Safety Whitepaper
Security Architecture Diagram PDF

Contact

For security questions or procurement reviews:

security@siftr.so