Data Loss Prevention (DLP)

Business objective

DLP prevents exfiltration of sensitive data via content- and lineage-aware controls. The seed’s metaphor: the bouncer at the exits — it stops sensitive data from walking out the door, now including via AI prompts and pastes into chatbots. Where dspm finds and labels data sitting still, DLP acts at the moment data tries to move — into an email, an upload, a USB drive, or a prompt to a public model.

The AI era reshaped DLP: the highest-risk egress path is now an employee pasting MNPI or client data into ChatGPT, or an agent sending data to an external tool. The newer “AI-native” DLP vendors focus on that prompt/clipboard surface; the incumbents focus on classic endpoint/network/email channels.

When you need it

Day-1. For a hedge fund the exfiltration of MNPI, client data, or IP is the nightmare scenario, and the cheapest place to stop it is at the exit. Classic DLP on email and endpoints is often already owned; the AI-specific gap — data leaving via prompts — is what is new and urgent the moment employees use public AI tools (they do). Pair it with ai-access-governance for prompt-aware enforcement.

Security role

A preventive, inline control on outbound data flows — it blocks exfiltration even when an attacker or careless user already has access to the data and an outbound channel lined up. Its core competence is recognizing what is sensitive in the outbound stream and stopping it in motion. It lives at the boundary between the green/yellow zones and the outside (red): the last gate before data crosses out of trusted territory.

Vendors

AI-native / prompt-aware DLP:

• cyberhaven — data-lineage-based DLP; tracks where data came from to judge egress; also ai-access-governance (data-flow visibility into AI usage).
• nightfall-ai — content-detection DLP with SaaS/AI coverage and API-based scanning.
• mind — AI-era DLP focused on data-in-motion.
• prompt-security — prompt-layer inspection and DLP for AI traffic (also ai-runtime-security, ai-spm); seed flags acquisition (see below).
• lasso-security — DLP/guardrails for LLM and agent traffic (also ai-runtime-security, agent-runtime-security).
• jazz-security — AI-native DLP (reclassified here from AI-SOC after research; it is a data-protection product, not a SOC analyst).
• audition-ai — in-tenant finance-vertical AI platform with built-in DLP (PII/SSN/PCI pattern matching) + behavioral “Generative Rules” applied to its assistant/agent traffic. Cross-listed; primary home enterprise-ai-assistant, also ai-governance-platform.

Platform / incumbent DLP:

• microsoft-purview — Purview DLP across M365/endpoint/cloud; default for Microsoft shops (also dspm).
• netskope — inline DLP within its SSE/SASE platform.
• forcepoint — long-standing enterprise DLP (endpoint/network/email).
• cyera — DSPM-led vendor extending into DLP/egress controls.

Consolidation / M&A dynamics

• prompt-security — acquired by SentinelOne (per seed; unverified — to confirm in research).

The pattern: AI-native DLP startups (Prompt Security, and the broader runtime/guardrail set) are being absorbed by endpoint, SSE, and runtime-security platforms that want a prompt-aware egress story; classic DLP increasingly ships inside SASE/SSE suites (Netskope, plus Palo Alto/Zscaler) rather than as standalone agents.

Adjacent categories

• dspm — supplies the classification DLP enforces on; the CSV merges them, the doc and this taxonomy keep them split (see Survey notes).
• ai-access-governance — overlaps heavily on the prompt-egress surface; “CASB for AI” is largely DLP plus shadow-AI discovery and intent awareness.
• ai-runtime-security — inspects prompts/responses at the model boundary; several DLP vendors (Prompt Security, Lasso) live here too.
• network-security-sase — the TLS-inspecting layer where inline DLP often runs.

Survey

Question. Which DLP / data-egress controls is your firm using or evaluating, including controls over data leaving via AI prompts?

Answer options. Cyberhaven, Nightfall AI, MIND, Prompt Security, Lasso Security, Microsoft Purview DLP, Netskope DLP, Forcepoint, Cyera, Other (Please Specify).

Response scale. multi-select; Interested; Considering/evaluating; Pilot/implementing; In production; Would recommend; Would not recommend.

Notes for survey design.

• The seed CSV asks one merged question (DSPM / Data Governance / DLP). Per taxonomy gap D2 we split it into dspm and DLP (this page). Vendors such as Cyera, Microsoft Purview, and Netskope answer both; expect the same vendor ticked on both pages, and reconcile that in analysis.
• Distinguish classic-channel DLP (Purview, Forcepoint, Netskope — often already owned) from AI-native prompt DLP (Cyberhaven, Nightfall, MIND, Prompt Security, Lasso). Respondents may not realize their incumbent suite already covers part of this.
• Overlap with ai-access-governance will confuse respondents — Cyberhaven and Lasso appear in both; consider asking “do you use a separate AI/prompt DLP control?” to disambiguate.
• M&A may date options: confirm Prompt Security (SentinelOne) status before fielding.

Open taxonomy questions

• The DLP / ai-access-governance boundary is the fuzziest in the data layer — prompt-aware DLP and CASB-for-AI are converging. Candidate for an explicit “where does egress control end and shadow-AI governance begin” note.
• Should incumbent suite-DLP (Purview, Netskope, Forcepoint) be surveyed separately from AI-native DLP? They solve overlapping but differently-scoped problems.
• CSV merge vs doc split (shared with dspm) is resolved in favor of the doc spine.