Data Exfiltration
Data exfiltration is the unauthorized transfer of data out of an environment to a place the owner does not control. It covers both deliberate theft by an attacker and accidental leakage by an insider. In AI deployments, a specific exfiltration channel has opened: sensitive data leaving in the prompts users and agents send to external model endpoints, and in the content those models return. An employee pasting customer records into a public chatbot, or a compromised agent routing internal data to a model API, is exfiltrating data over the AI request path.
How AI creates a new exfiltration channel
Traditional data-loss controls watch email, file transfers, and web uploads. An AI prompt is a different channel: it looks like ordinary HTTPS traffic to a model endpoint, and network-layer tools sitting under TLS cannot read the prompt content inside it. So PHI, source code, customer PII, or board-level financial detail can flow into a vendor LLM inside a normal-looking request, without tripping controls built for the older channels.
How the AI exfiltration path gets closed
Closing this channel means reading the prompt payload itself, classifying what it contains, and enforcing policy before the request reaches the model. DeepInspect terminates the AI request as a stateless proxy, applies prompt-level classification and per-identity policy, and either redacts, blocks, or allows the call based on what the payload holds and who is sending it. Every decision produces an audit record bound to the identity, so a blocked exfiltration attempt and an allowed benign request are both accounted for on the HTTP AI path.
Related reading
- Data Exfiltration via LLM: The Channel Your DLP Was Never Built to See
An LLM prompt is an HTTPS POST to a provider API carrying whatever the user or agent put in the context window. That makes it an egress channel with no size limit, no content inspection, and no identity correlation in most enterprises. This walks the four exfiltration paths through an LLM (deliberate paste, agent retrieval, injected instruction, response-side leakage), what each looks like on the wire, and the control that closes them.
- LLM Egress Control: The Per-Request Identity, Classification, and Audit Layer for AI Provider Traffic
LLM egress control is the request-time enforcement layer between corporate applications (and agents) and the external LLM endpoints they call. The layer reads the identity the request carries, classifies the prompt body, evaluates per-route policy, applies a pass, modify, redact, or block decision, and commits a per-decision audit record. This piece walks through the egress surface the layer covers, the policy decisions the layer commits, the audit record format, and the deployment topology that handles single-region and multi-region traffic.