Panoptes is a data-loss-prevention tool for AI assistants. It inspects what you send
to AI chat services and warns you when it detects sensitive data (PHI, PII, credentials,
financial data) before you send it. Because Panoptes exists to protect sensitive data, it
is built to hold as little of it as possible and to send none of it anywhere you did not
explicitly configure.
This policy is the privacy policy for the Panoptes browser extension
(Chrome / Chromium), published by Ironwright. Panoptes is also available as a separate VS
Code extension with additional enterprise DLP, SIEM, and email alert integrations; those
integrations do not exist in the browser extension and are called out below wherever they
would otherwise be confused with it.
The short version
Detection happens on your device. Panoptes uses no AI and makes no
network request to analyze your text.
Panoptes collects no analytics and no telemetry. There is no usage
tracking, no call-home on install or update, and no hidden endpoint.
By default, nothing leaves your machine. Every alert channel is
opt-in and starts empty.
When you turn on an alert channel, Panoptes sends redacted metadata about a
detection, never your raw prompt text.
What Panoptes processes
Prompt and response text on supported AI-service pages, in memory,
in order to scan it. This text is not persisted and is not transmitted as part of
scanning.
Detection results: the type of sensitive data found, a masked
snippet of the matched value, its position, and counts.
Optional context for alerts that you enable: the AI service name,
the page URL, basic browser/machine information, and a username. This browser
extension does not request the browser permission needed to read your profile
identity, so the username field is always sent as "unknown-user." It does not capture
Active Directory or Entra ID identities; that directory-identity capability exists
only in the separate Panoptes VS Code extension, not in this browser extension.
Two capabilities in the browser extension are off by default and
process more than text once you turn them on:
Images you attach to a prompt (setting: "Scan images", off by
default; on ChatGPT and Claude.ai only). Sensitive data pasted as a screenshot is
invisible to a text scanner, so Panoptes can read text out of an attached image using
optical character recognition and scan that text. The OCR engine is bundled
inside the extension and runs on your device: the image is never uploaded,
not to Ironwright and not to your own alert destinations. The text recovered from the
image is the sensitive data, so it is never included in an alert either. An image
detection sends the same redacted metadata as a text detection, plus two numbers
describing how legible the image was and a count of how many characters of text were
recovered from it. Image scanning only warns; it never blocks an upload.
An AI provider's model-training setting (setting: "Audit training
opt-in", off by default; on ChatGPT and Claude.ai only). Panoptes reads whether that
provider's own "improve the model for everyone" switch is on, so it can tell you when
your conversations are being retained for training. It reads only that switch's
state, never changes it, and reads no other part of your account.
What Panoptes sends, and where
This browser extension only sends data to destinations you configure.
Out of the box, all three of these are off:
Channel
Default
What is sent
Webhook (Slack/Teams/custom endpoint)
Off
Redacted finding metadata
Central reporting server
Off
Redacted finding metadata
Policy server (organization deployments)
Off
Nothing about you or your text. Panoptes only fetches your organization's detection policy, sending the access token you configured and, if set, your team name
This browser extension has no SIEM, DLP, or email (SMTP) alert channel. Those
integrations (Splunk, Elastic, QRadar, Sentinel, Chronicle for SIEM; Purview, Symantec,
Forcepoint, Digital Guardian for DLP; and SMTP email) exist only in the separate Panoptes
VS Code extension, not in this browser extension.
Before any alert is sent, Panoptes strips the raw prompt/response snippet and masks
matched values, so the sensitive data that triggered the alert is not reproduced in the
alert. Every destination above must use https, whether or not it carries a
credential; Panoptes refuses to send to a plain http endpoint.
What Panoptes stores
Your settings (which services to monitor, detection options, and
any alert channel configuration you enter), in the extension's local settings
storage.
A cached copy of the enforcement policy, and a note of whether the
last fetch succeeded, if your organization uses a policy server.
A short record of alerts already sent, used only to avoid alerting
you twice about the same thing. If you turn on the training opt-in audit, this
records the AI service name and the time of the last alert for it. This never leaves
your machine.
Housekeeping values the extension needs to run: a timestamp that
keeps its background worker alive, and a flag recording whether the toolbar icon is
currently showing an alert.
Panoptes does not store your prompt text, and does not store
detected values in the clear.
A local detection-metadata log file is a feature of the separate Panoptes VS Code
extension, not this browser extension; the browser extension writes no log file to
disk.
Third parties
Panoptes has no advertising, no third-party analytics, and no data brokers. The only
third parties that ever receive data are the alert destinations you point
this browser extension at (your own webhook or your own central monitoring server).
Your control
Everything that leaves the machine is opt-in; remove a channel's configuration and
nothing is sent there.
Uninstalling the extension removes it and its settings storage.