Perplexity Data Loss Prevention
Redact sensitive data in every Perplexity query - web, mobile apps, the API, and Comet's agent mode. In-flight redaction that runs before Perplexity sees a thing.
14-day free trial, no credit card.
Summarise this customer complaint: "Ravi Mehta (ravi.m@finovate.in[REDACTED_PERSONAL_INFO_1], +91 98201 22334[REDACTED_PERSONAL_INFO_2]) says his card 4539 1488 0343 6467[REDACTED_FINANCIAL_DATA_1] was charged twice. Account #FNV-880221, support ticket SUP-9911. Our prod webhook is at https://api.finovate.in/hooks with HMAC_SECRET=fno_live_8ZqQ12pMxRn[REDACTED_CREDENTIALS_1]."
What leaks to Perplexity - and why
The four exfiltration patterns we see most often when teams adopt Perplexity.
Free, Pro, and Max plans train on your queries by default
On the consumer tiers, Perplexity uses your prompts to improve its models unless you explicitly toggle off "AI data retention" in Account Settings - Preferences. Most employees never find that switch. Only Enterprise Pro and Enterprise Max ship with a contractual "Zero Data Training" guarantee.
Comet's agent mode reads everything on the page
Comet is Perplexity's Chromium-based AI browser with an in-page assistant that can act across tabs - shopping, drafting emails, summarising content. To do that it pulls the visible DOM of the current page into context. Open a CRM record, an internal dashboard, or a draft email and the assistant is reading it.
Spaces remember context across multiple tabs
Comet Spaces (workspaces) group related tabs so the assistant has full context across all of them. One sensitive tab pollutes the entire space - the assistant can quote from a customer record while you ask an unrelated question in a different tab.
"Summarise this for me" prompts leak internal documents
Perplexity's research framing invites users to paste long-form content - contracts, meeting notes, draft RFCs, support transcripts - and ask for a summary or counter-argument. The whole document hits Perplexity's logs before the summary comes back.
Voice mode and follow-ups echo prior turns
Voice mode (powered by an OpenAI Realtime model) and chained follow-up queries carry the conversation history forward. A sensitive turn early in the session shows up in every subsequent request as context.
What Perplexity actually sees, with NexusNest in front
The user types whatever they want. NexusNest redacts the sensitive spans in-flight, so the prompt that reaches Perplexity AI has placeholders in place of the secrets.
Summarise this customer complaint: "Ravi Mehta (ravi.m@finovate.in, +91 98201 22334) says his card 4539 1488 0343 6467 was charged twice. Account #FNV-880221, support ticket SUP-9911. Our prod webhook is at https://api.finovate.in/hooks with HMAC_SECRET=fno_live_8ZqQ12pMxRn."
Summarise this customer complaint: "Ravi Mehta ([REDACTED_PERSONAL_INFO_1], [REDACTED_PERSONAL_INFO_2]) says his card [REDACTED_FINANCIAL_DATA_1] was charged twice. Account #FNV-880221, support ticket SUP-9911. Our prod webhook is at https://api.finovate.in/hooks with [REDACTED_CREDENTIALS_1]."
Set up in 2 minutes
Install the agent
Download the .pkg / .exe and double-click. The agent installs a local trusted CA and the system proxy - no IT ticket required for Perplexity traffic to flow through it.
Open your AI tool as normal
Use Perplexity exactly the way you do today - browser, desktop app, or API. The agent intercepts the outgoing request, runs the redaction pipeline, and forwards a redacted version.
Watch the dashboard
Every prompt shows up in the admin dashboard with what was redacted, by which employee, on which machine. Perplexity usage becomes legible.
Perplexity DLP - common questions
Does NexusNest work with the Comet browser?
Yes. Comet is Chromium-based and routes its AI requests over HTTPS like any other client. The agent installs a system proxy so prompts from Comet's assistant - including the page context it ingests for agent tasks - are scanned and redacted before they reach Perplexity.
Doesn't Perplexity Enterprise Pro already cover this?
Enterprise Pro adds Zero Data Training, SOC 2 Type II, SAML 2.0 SSO, audit logs, and configurable retention. None of that redacts what users type. Your prompts and the pages Comet reads still leave the device verbatim - Perplexity just promises not to train on them. NexusNest removes the sensitive content before any provider sees it.
Can it stop the Free / Pro / Max default-training problem?
Yes - that's exactly the gap it fills. Even if individual employees forget to flip the "AI data retention" toggle, the redaction layer means the data Perplexity receives never contains your customers' identifiers, secrets, or internal text in the first place.
Does redaction affect Comet's agent tasks - shopping, email drafting, etc.?
Only sensitive content is redacted, not the structural context. Shopping queries, navigation, and generic page content pass through. Customer PII, secrets, and internal-doc paragraphs are replaced with opaque placeholders that the assistant treats as normal tokens.
What about the citations Perplexity adds to its answers?
Perplexity's citations are sourced from the public web, not your prompt. Redacting your input does not change which sources are cited - it only prevents your sensitive data from joining the upstream payload.
Stop sensitive data leaking to Perplexity today
Deploy on every employee laptop in under 10 minutes. 14-day free trial. No credit card required.