Criterion
How it worksDecrypts traffic on the network path (MITM), inspects it and re-encrypts it. Uniquely, it can mask ID and card-number patterns in a prompt instead of blocking it.Injects code into the app's memory to hook its encryption functions, reading plaintext before TLS or certificate pinning ever runs. Allow or block only.An OS-level agent that judges uploads, copies and pastes by the source file: its Purview tag, or ID/card-number patterns in its content. Block, warn or audit; no masking.Traces clipboard and file events back to a sensitive source, without reading content. It's also what switches off an Electron app's certificate pinning so a network proxy can read its traffic. Allow or block only.Reads the prompt's own meaning, and what it's asking the AI to do, on-device before it's sent; no tag or file event required. Allow or block only.Coding agents are pointed at it instead of the vendor. It reads every request as plain text, masks or blocks ID and card-number patterns, and sends the rest on under the org's own key.
Detection method
User experience
Upkeep
Price (per user / month)$3 – $10$10 – $25$15 – $30$15 – $30$15 – $30—
Coverage gaps
  • Pinned desktop apps

    they reject its certificate, so it can't read their traffic

  • Files an agent reads on an enterprise account

    it's sanctioned use, so the proxy lets it through without reading it

  • Uploaded proprietary files when untagged

    there's no ID or card-number pattern to match

  • Pasted proprietary text and proprietary text typed from memory

    there's no ID or card-number pattern to match

  • Screenshots

    an image has no text to scan and no link back to its source

  • Anything on macOS

    macOS blocks the memory hook it depends on

  • Proprietary content

    there's no ID or card-number pattern to match

  • Screenshots

    an image has no text to scan and no link back to its source

  • Pasted proprietary text, uploaded proprietary files and proprietary files an agent reads when untagged

    there's no Purview tag and no ID or card-number pattern

  • Screenshots

    an image has no text to scan and no link back to its source

  • Typed text

    typing isn't a copy, paste or upload, so there's no device action to judge

  • Pasted text on macOS

    macOS restricts clipboard tracking

  • Sources outside its graph

    it has nothing to trace them back to

  • Screenshots

    an image has no text to scan and no link back to its source

  • Typed text

    nothing was copied, so there's no event to trace

  • Files an agent reads

    the agent reads them itself, with no paste or drop to trace

  • Uploaded files

    it reads prompts, and a whole file isn't one

  • Screenshots

    an image has no text to scan and no link back to its source

  • Files an agent reads

    it reads what's typed, and the agent attaches the files itself

  • Browsers and desktop apps

    they go straight to their vendor and can't be pointed at a gateway

  • Personal keys and logins

    they go straight to the vendor's API and never reach the gateway

  • Proprietary content

    there's no ID or card-number pattern to match

  • Screenshots

    an image has no text to scan and no link back to its source

Compliance helped