Perplexity
Perplexity AI, Inc.
Short answer
Short answer: Perplexity is reasonable for public research, and safe for work on Enterprise Pro (DPA, SOC 2, no training on your data). On the Free or Pro tier, keep personal and client data out of your queries.
At a glance
- Vendor / legal entity
- Perplexity AI, Inc.
- Headquarters
- 🇺🇸 San Francisco, US
- Category
- LLM / Chat
- Ownership / jurisdiction
- US company, headquartered in San Francisco.
- Trains on your data?
- On by default — opt-out available
- Data hosting & residency
- US-based processing. A verified EU data-residency option could not be confirmed from primary sources.
- Sub-processors
- Sub-processors (incl. AWS, Microsoft Azure) are published in the Perplexity Trust Center.
- Enterprise tier
- Enterprise Pro excludes your data from training, adds SOC 2 Type II, a DPA, custom retention and workspace/admin controls; the Sonar API is zero-retention. The Free/Pro consumer tiers offer only a data-retention toggle.
- Certifications
- SOC 2 Type II · HIPAA gap assessment (2025) · DPA available (Enterprise)
Why this rating
Consumer Free/Pro/Max use your data for training by default (an AI data-retention toggle turns it off, future-only). Enterprise Pro excludes training, holds SOC 2 Type II and adds retention controls, and the Sonar API is zero-retention. A verified EU data-residency option could not be confirmed. Medium: fine for public research, governed on Enterprise Pro, exposed on the consumer tiers.
What it is
Perplexity combines an LLM with live web search and inline citations. Because people paste work context into their questions, the concern is the same as any consumer LLM: on Free/Pro/Max your data feeds training unless you disable AI data retention, and there is no consumer DPA. Enterprise Pro is the tier to standardise on for regulated teams; the Sonar API is zero-retention.
Legal & compliance examples
How the same tool can be a safe helper or a high-risk deployment — the difference is what it decides about a person. Examples authored with Qadar AI's governance findings.
Don't: Screen candidate CVs and auto-reject applicants, or infer a candidate's emotions in a video interview.
Fine: Draft a job description, summarise public market research, or brainstorm interview topics — with no automated decision about a person.
These examples are general information, not legal advice, and are pending a legal review. Your obligations depend on your exact use, tools, and set-up.
Frequently asked questions
Is Perplexity private and safe for work?
On Enterprise Pro, yes — it excludes your data from training and provides a DPA and SOC 2 Type II. On the consumer tiers, treat your queries as leaving the organisation and avoid personal or client data.
Does Perplexity train on your data?
The consumer tiers use your data to improve the service unless you turn off the AI data-retention toggle. Enterprise Pro and the Sonar API exclude it.
How do I govern Perplexity at work?
Qadar AI Shield discovers Perplexity usage, redacts personal data before it reaches the query, and lets you allow-list or block it per group.
Sources
Every fact and the risk rating on this page trace to a primary source — the vendor's own trust center, privacy policy, DPA, or a regulator. Verify current terms before relying on them.
Govern this tool
Qadar AI Shield turns these risk facts into enforcement — discover who uses the tool, redact personal data before it reaches it, and allow-list or block it per team.
Risk levels are Qadar AI's curated assessment from cited primary sources — a curated assessment, not a certification, and not legal advice. Verify a tool's current terms before relying on them.