AI Governance & Security Vendor Wiki

7 min read

Perplexity Enterprise

Researched 2026-06-28. Primary category: enterprise-ai-assistant.

One-liner — An enterprise “answer engine”: a web-search-grounded AI assistant with citations, plus enterprise add-ons (Spaces, internal file/app connectors, SSO/SCIM, admin controls, no-training guarantee).

What it does

Perplexity is primarily a web-search assistant / answer engine — you ask a question, it searches the live web, synthesizes an answer, and shows inline citations. Enterprise Pro and Enterprise Max wrap that consumer product in a governance shell: it stops training on your data, adds Spaces (shared workspaces / collections), and lets the assistant search across internal files and connected work apps (Google Drive, OneDrive, SharePoint, Box, Dropbox) alongside the web. The pitch to a knowledge-work org is “Google + ChatGPT for research,” with sources attached so analysts can verify claims.

Note the center of gravity differs from openai-chatgpt-enterprise / anthropic-claude-enterprise: Perplexity leads with external web research and citations, not general-purpose chat or coding. Internal-knowledge RAG is a newer add-on, not the core.

Where it sits in the stack

  • Category: enterprise-ai-assistant (UX layer) — the chatbot end-users actually touch.
  • Risk exposure: as a research assistant it pulls untrusted web content into the model (prompt-injection surface), can be pointed at sensitive internal data via connectors, and has an inherent egress path (it sends queries to the web and to third-party model providers). All three risks are in play at once, which is exactly why the enterprise controls below matter.
  • It is a consumer of, not a substitute for, perimeter controls like an ai-gateway, ai-access-governance (shadow-AI / CASB-for-AI), and dlp. A fund would typically still want shadow-AI discovery to see who’s using the consumer Perplexity vs. the sanctioned enterprise tenant.

Deployment & architecture

  • SaaS only. No self-hosted / on-prem option; runs in Perplexity’s cloud. Models are a mix of Perplexity’s own and third-party frontier models (OpenAI, Anthropic, Google) accessed under contract.
  • Identity: SAML 2.0 SSO (Okta and other IdPs), SCIM provisioning (Okta, Entra ID, Google Workspace), MFA, short-lived session credentials.
  • Connectors: Google Drive, OneDrive, SharePoint, Box, Dropbox. Connectors are permission-aware at view time — e.g. a file deleted or de-permissioned in SharePoint is immediately removed from Perplexity; files surfaced from connected apps in a Space are searchable by anyone with Space access, but a user needs rights on the underlying app to see file contents. Admins enable/disable each connector org-wide.
  • Admin & audit: central admin panel governs file upload/download, sharing of Threads/Pages/Spaces, and connectors. Audit Logs capture end-to-end queries (input, agent steps, answers) plus admin setting changes (event type, timestamp, user email, IP) and are pushed in real time to a customer-configured webhook (HTTP POST) — that is the main hook into a SIEM/SOC. Audit Logs are gated to orgs with 50+ seats or ≥1 Enterprise Max user. Data export in JSON/CSV/PDF.

Positioning & differentiators

  • Differentiator: citation-first web research. For an investment shop the appeal is fast, sourced answers on markets, companies, filings, and news — with links to chase down.
  • Governance posture (the CTO-relevant part):
    • No training on enterprise data — Enterprise content is not used to train Perplexity’s or third-party providers’ models; Perplexity says it holds contractual no-train terms with model providers, reviewed annually.
    • Retention controls — uploaded files default to short retention (~7 days; some materials cite 1-day auto-delete); the containing Thread persists until deleted. Orgs with 50+ Pro seats or ≥1 Max seat can set custom retention and force-delete on demand.
    • Certifications — SOC 2 Type II (2026 attestation), HIPAA-aligned safeguards (HIPAA gap assessment 2025), GDPR, plus a Trust Center (trust.perplexity.ai) and CAIQlite. DPA available. PCI DSS appears in marketing summaries — verify before relying on it.
    • Data residency — no evidence of configurable regional residency (e.g. EU-only) for the enterprise assistant; treat as not available until confirmed. This is a gap vs. some Microsoft/Google offerings.
  • vs. neighbors: glean is the closer comparison for internal entitlement-aware enterprise search/RAG (entitlement-aware-rag); Perplexity’s connectors are newer and web-research-led. openai-chatgpt-enterprise and anthropic-claude-enterprise are stronger general assistants; microsoft-365-copilot / gemini-enterprise win on native data-estate integration and residency.

Ownership, funding & M&A

  • Independent, private, venture-backed. No seed M&A flag and none found — Perplexity is an acquirer-of-talent/none-of-note, not acquired. Ownership confidence high (well-documented private company).
  • Valued ~$20B (Sept 2025) after a ~$200M round; reports of a Series E-6 ~$21B valuation early 2026 (figures vary $18–21B across outlets). Investors: Nvidia, Jeff Bezos (personal), SoftBank Vision Fund 2, IVP, Accel, NEA, Databricks, Bessemer, Nat Friedman, Tobias Lütke. Microsoft committed ~$750M GPU capacity (Jan 2026).
  • Funding total is uncertain: Wikipedia-disclosed rounds sum ~$665M; aggregators (Tracxn/TexAu) report ~$1B–$1.7B. Left as a range, not a point estimate.

CTO / hedge-fund lens

  • Day-1 question, conditional fit. As an answer engine it’s a genuinely useful research tool for analysts and is cheap to pilot. The governance shell (no-train, SSO/SCIM, audit logs, retention controls) clears the basic enterprise bar.
  • Where a fund should be careful:
    • Comms surveillance / eDiscovery: there is no native archival or eDiscovery/WORM hook for MAR/MNPI-style supervision. The audit-log webhook can stream activity to a SIEM, but that is monitoring telemetry, not a compliant comms archive. If analyst interactions need to be captured for surveillance (comms-surveillance) or legal hold, you’d have to build it off the webhook/export — flag this explicitly with Compliance.
    • Data residency appears unavailable — a problem for EU/region-locked mandates.
    • SaaS-only + third-party model providers means data leaves your tenant to multiple providers under contract; acceptable for many, but diligence the subprocessor list.
  • SR 11-7 / model risk: low direct relevance — it’s an assistant, not a model you deploy in a decision pipeline. Treat outputs as un-validated research, not model output.
  • Fit: medium. Good sanctioned research tool; not a system of record, not a surveillance-grade comms channel. Most useful where citations and live web matter more than deep internal-data RAG.

Competitors / alternatives

Open questions / to verify

  • Exact default retention (7-day vs 1-day) and whether custom retention covers Threads/Pages, not just uploaded files.
  • Whether data residency (e.g. EU) is offered at all for the enterprise assistant.
  • Current subprocessor / model-provider list and PCI DSS scope (marketing claim).
  • Precise cumulative funding figure and confirmation of the early-2026 Series E-6 valuation from a primary source.

Sources

History

  • [2026-06-28] Stub created from seed registry.
  • [2026-06-28] Researched; established Perplexity Enterprise as SaaS web-search answer engine with enterprise governance shell (no-train on enterprise data, SAML SSO + SCIM, MFA, admin/audit controls with real-time audit-log webhook, 7-day default file retention + custom retention for larger orgs, permission-aware connectors), SOC 2 Type II / HIPAA / GDPR certs; ownership independent/private venture-backed ($20B Sept 2025, high confidence) with no M&A; flagged absence of native eDiscovery/comms-archival and apparent lack of data residency as hedge-fund gaps. fit = medium.

Graph View

Backlinks