Perplexity

Models & services in use

No model dependency.

Vendor due-diligence questionnaire

Auto-populated from public disclosures · gaps flagged for human input

• Data residency commitments documented
  Yes — us-east-1, eu-central-1 (Anthropic Workspace)
• EU AI Act provider obligations published
  Yes — public Trust Center
• DORA ICT third-party documentation
  On file — DPA addendum signed Mar 2026
• FREE-AI BFSI vendor disclosure
  Auto-collected — no material gaps
• Sub-processor list available
  Yes — published quarterly
• Model deprecation policy
  12-month notice; LTS releases identified
• Incident notification SLA
  ≤ 24h for material incidents
• Data isolation guarantees
  Per-workspace encryption, no training on customer data

Trust Lab-conducted assessments

• Pre-approval eval — claude-3-5-sonnet

  FinanceBench 78.4% · IndicSafe 91.2%
  Pass
• Pre-approval eval — claude-3-5-haiku

  FinanceBench 71.8%
  Pass
• Frontier red-team — claude-3-5-sonnet (OWASP+ATLAS)

  12 successful probes; mitigated via prompt + RAG hardening
  Pass with mitigations
• Capability evaluation — claude-3-7-sonnet (in eval)

  Pre-approval policy v4 active
  In progress

Continuous monitoring

• Apr 9 · Anthropic published updated sub-processor list — auto-diff: no material change.
• Mar 28 · claude-3-7-sonnet released — pre-approval policy v4 auto-triggered.
• Mar 18 · Quarterly assessment refresh — Approved through Q3 2026.