In May, Netskope announced an integration with the Claude Compliance API, giving security teams visibility into Claude Enterprise activity, DLP-aligned policy enforcement on uploaded and generated files, and continuous compliance monitoring across frameworks such as GDPR, HIPAA, and PCI-DSS. That integration answered a critical first question for enterprises adopting Claude at scale: What is happening inside our AI environment, and is it compliant?
The next question is arguably harder, but no less important: Can we stop sensitive data from reaching the model in the first place?
And so Netskope is now expanding our integration with Anthropic’s Claude to answer that question with a native DLP validation layer, delivered through Claude’s inference hooks protocol, inspecting prompts and the data from the tool calls before they reach inference.
Why “prompt-time inspection” matters
Once sensitive data enters an LLM’s inference layer, it’s effectively baked in: Traditional deletion and redaction don’t apply to AI knowledge banks the way they do for a file sitting in storage. Once a model ingests information through a prompt, it’s almost impossible to make it forget, creating a risk that one ill-thought through AI query could create an invisible and ongoing risk of sensitive, confidential or regulated data exposure. That makes prompt-time inspection (as opposed to after-the-fact monitoring) critical for data security AI use cases. Compliance visibility tells you what happened; a validation layer in the prompt path decides what’s allowed to happen.
It is this architectural gap that the extended integration is designed to close: Security teams get a real-time control point in front of every user prompt, workload, and tool call, not just a record of it afterward.
How it works
The integration uses a hook pre-inference: Claude calls out to a webhook before a prompt reaches the model, and Netskope One DLP On Demand returns an allow/deny verdict based on the customer’s existing DLP profiles.
- Claude Enterprise surfaces (Claude chat, Claude Cowork, Claude Code and Claude Design) route prompts and the data from the tool calls (including MCP, skills, and plugin traffic) through the hook.
- Netskope translates the generic webhook payload into a DLP On Demand inspection request, scanning against whichever profiles (PII, PCI, PHI, IP, etc. including custom DLP profiles) the customer has deployed in their cloud or on-prem infrastructure.
- The verdict (allow or deny, with a reason) returns before inference happens, and blocked prompts generate an incident for incident response and forensic analysis or end-user feedback rather than silently failing.
Because it’s built on a webhook protocol rather than an agent or proxy, there’s no new infrastructure to deploy and no meaningful latency, a design goal carried over directly from the Compliance API integration but with the benefit of moving the data security verdict-action inline.
What this adds for existing Netskope customers
| Capability | Compliance API integration (May 2026) | DLP Validation Layer / Prompt Hooks (new) |
|---|---|---|
| Where it acts | After the fact: activity, files, and configuration surfaced into Netskope | Before inference: prompts and the data from the tool calls inspected in real time |
| Primary value | Visibility, audit-readiness, compliance mapping | Prevention: block restricted data before the model ever sees it |
| Coverage | Claude Enterprise activity, files | Claude chat, Claude Cowork, Claude Code and Claude Design prompts and tool calls from any integration surface |
| Enforcement | Policy applied to files already in the environment | Existing DLP profiles extended natively into the prompt path with the benefit of real-time user feedback/coaching |
Together, the two integrations cover the full lifecycle: govern what’s already there, and prevent what shouldn’t get in.
This capability is currently supported by Netskope DLP in private beta. Public beta and general availability are targeted for later in the year.
Request a demo today or please get in contact with your Netskope contact to participate in the beta.
