Zero-trust network access usually carries a deployment tax: an agent on every device that needs to reach private applications. Agents are friction — they fail to install, lag on updates, and simply cannot run on unmanaged or third-party machines. Agentless access removes that tax for the cases agents cannot cover.

Zscaler, Inc.'s US11811855B1, “Policy based agentless file transfer in zero trust private networks” (issued November 7, 2023; CPC H04L 67/06 — file transfer over networks, and H04L 67/02 — HTTP-based protocols), describes agentless, policy-governed file transfer inside a zero-trust private network. Read it at US11811855B1.

“Systems and methods for policy based agentless file transfer in zero trust private networks.”— U.S. Patent No. 11,811,855 source

The problem the filing names is precise. Existing zero-trust file transfer “either require[s] an agent installed at the end point, or a silo application that is opaque to zero trust policies” — and an installed agent “do[es] not work well with use cases where the endpoint user is a third party and not in a long-term relationship with the customer.” The proposed fix is architectural: “extending remote pixel rendering protocols to support file transfer from the endpoint.” The user works through “an agentless, browser-based interface” — a clientless remote-desktop gateway tunnel — so the file moves over standard web protocols with nothing installed on the device.

Policy is evaluated per request, not per device trust. Claim 1 has the system “receiving a request for a file transfer; determining a file transfer protocol; evaluating one or more criteria associated with the request,” where those criteria include “policy associated with the end user and file inspection results.” The specification lists the inputs feeding that decision: “Security Assertion Markup Language (SAML) attributes, client posture, end user network.” The worked example is asymmetric on purpose — “a trusted user coming from a trusted network may be assigned access to both download and upload,” while “unknown temporary users… from an untrusted network” may get upload-only rights. The trust decision lives in the broker at the private access service edge, not in an agent on the machine.

The file-inspection path is the security guarantee that makes agentless transfer acceptable. Per claim 1, “responsive to an end user's policy including a requirement for file inspection,” the system “send[s] the file to a sandbox for inspection” and “receiv[es] a result of the inspection from the sandbox.” The specification adds the operational detail: an “exporter” proxies the upload and holds the file in a cache “until a sandbox report comes back with a verdict,” and if no verdict arrives within a predefined window — it gives 10 minutes as an example — “the upload operation fails, and the file is removed.” Underlying protocols are mapped to console type: SFTP for SSH/VNC consoles, RDPRD for RDP. Agentless, in other words, does not mean uninspected; the broker still detonates the file before letting it land.

The dependent claims confirm the agentless, browser-first posture and the inspection guardrails. The request “is received from a browser-based interface associated with an endpoint user” (claim 2), and the file-transfer protocol is “determined based on a type of browser-based interface used” (claim 3) — the protocol adapts to whichever clientless console the user is on. The inspection results can draw on “a historical file assessment to identify a file as malicious or benign” (claim 4), so a previously seen file need not be re-detonated. And the fail-safe is explicit in the claims, not just the specification: the transfer “is failed responsive to the file inspection taking longer than a preconfigured amount of time” (claim 5), with the file “held in a cache during the file inspection” (claim 6). The default is deny: if the sandbox cannot return a verdict in time, the upload does not happen.

The specification is careful about the failure and abuse cases, which is where agentless transfer usually goes wrong. To enable the clientless gateway transfer, “exporters enable Secure File Transfer Protocol (SFTP) or Remote Desktop Protocol” transfer “depending on the type of the console” — SFTP for SSH or VNC consoles, RDP-based transfer for RDP. Files identified as benign “are uploaded and then removed from the exporter's file cache,” while files identified as malicious “are immediately deleted and the browser is informed of the threat.” The design explicitly aims to “protect customers content in case of an exporter malfunction and protect exporter resources from exhaustion due to customers files in the cache” — so when inspection is off, content is never cached at all, and when it is on, the cache is transient and bounded by the verdict timeout. In other words, the convenience of browser-only access is fenced by a broker that cleans up after itself and fails closed, which is what lets a security team extend private access to a contractor's unmanaged laptop without expanding its attack surface.

Why this is a business story: agentless reach is a competitive wedge in the ZTNA market, where Zscaler, Palo Alto's Prisma Access, and Cloudflare fight over coverage of contractors, BYOD, and partner access. The strategic value is total addressable use cases — every browser-only, third-party, or unmanaged-device scenario an agent cannot cover is revenue a pure-agent competitor leaves on the table, and the SAML-attributes-plus-posture policy model is what lets a vendor say yes to those users without weakening the trust story. The long inventor list on this grant signals a substantial internal effort behind the capability.

The grounded read: agentless zero-trust transfer brokers files through a clientless, browser-based gateway over standard web protocols, decides access per request from SAML attributes and posture, and detonates uploads in a sandbox before they land — extending secure private access to devices that cannot run an agent. Zscaler's 2023 grant names that capability with its broker-side enforcement and sandbox-verdict caching — a coverage wedge in the contested ZTNA market.