Yes — on company-managed devices. CircleScope is a browser extension deployed by IT policy to managed Chrome and Edge. It captures prompts submitted to ChatGPT, Claude, Gemini, Microsoft Copilot, M365 Copilot, and Perplexity, scores each one for business risk with AI, and alerts your team when something needs attention. It only watches the AI sites — never browsing history, keystrokes, or other tabs.
AI prompt monitoring is visibility into what employees actually submit to AI tools — the prompt text and file uploads — so an organization can detect sensitive data leaving (customer records, financials, source code), coach the people involved, and produce evidence that its AI acceptable-use policy is followed. It differs from web filtering (which only sees the domain) and from blocking (which just pushes usage to personal devices).
Yes — and without installing anything. One admin consent connects your Microsoft 365 tenant to CircleScope, read-only. From then on, Copilot activity inside Teams, Outlook, Word, Excel, the Edge sidebar, and the Copilot app is recorded server-side from Microsoft’s own audit log: who used it, where, whether it searched the web, and which files it read. Microsoft’s own jailbreak and prompt-injection flags come through too, so you hear about an attack aimed at Copilot itself. For users licensed for Microsoft 365 Copilot, the prompt text is captured too and scored like any other prompt. Because it is server-side, it covers phones and unmanaged devices that a browser extension never reaches. Copilot’s answers are never collected.
No. The two are complementary and cover different directions of risk. The extension watches data going out — what employees type or upload into public AI sites like ChatGPT and Claude from managed browsers. The Microsoft 365 connection watches what Microsoft’s own AI does inside your tenant — which documents Copilot read to answer, and whether they were labeled — plus which third-party apps have been granted access to your data. Most organizations want both; either one works on its own.
They are the sleeper risk: they do not run in a browser tab, so no extension sees them. They join meetings and read calendars through a permission an employee granted once — the “Allow this app to access your calendar” screen — and keep it indefinitely. CircleScope’s connected-apps audit reads those grants directly from your Microsoft 365 tenant every day, matches them against a maintained catalog of AI vendors, and shows exactly what each app can reach and who consented — so you can approve the ones you use on purpose and revoke the rest.
Every Business and Enterprise plan — Business Basic, Standard, Premium, E3, E5 — gets Copilot activity monitoring and the connected-apps audit, because both rely on audit logging Microsoft enables by default. The one thing licensing gates is prompt text: Microsoft only exposes Copilot prompt content for users holding the Microsoft 365 Copilot add-on. Free Copilot Chat users are still monitored — who, when, where, and which files were read — just without the words typed. The portal shows this per user, so you always know which kind of visibility you have for whom. One optional detail — when each connected app last accessed your data and who approved it — needs Business Premium or Entra ID P1; on Business Standard the portal says so instead of leaving the column blank.
Shadow AI is employee use of AI tools that IT has not approved or does not know about — new chatbots, AI coding assistants, character apps, and niche services that appear faster than any blocklist. CircleScope detects visits to a maintained watchlist of known AI services (by domain only, never page content), so you learn which new tools your people are adopting before they become a problem.
In practice, no. Block the sites and employees switch to phones and personal accounts — the organization loses the productivity gains and the visibility at the same time. Monitoring with coaching keeps the tools available, catches the genuinely risky moments, and changes behavior through documented conversations instead of bans.
CircleScope is deployed only on company-managed devices under the organization’s acceptable-use policy, and we ship disclosure language so employees are properly informed. It captures only AI-site interactions — no keystrokes, screenshots, passwords, camera, or general browsing — and sensitive values like Social Security numbers are redacted on the device before anything is transmitted. Employers should confirm requirements for their jurisdiction.
Cyber-insurance applications and client security questionnaires increasingly ask, in writing, how AI use is governed: whether employees can paste sensitive data into AI tools, whether usage is monitored, and whether incidents are documented. "We monitor AI use, coach on it, and here is the report" is a fundable answer — a policy PDF alone is not.
Traditional DLP suites are heavyweight, expensive, and not AI-aware — they were built for email and file shares, not for prompts. CircleScope is purpose-built for the AI surface: it understands prompt context, classifies risk with AI (PII, PHI, credentials, source code, financial, HR, legal), writes incident summaries, and pairs detection with a coaching workflow. It deploys in minutes to the browser employees already use.
Collected: prompt text submitted to monitored AI services, file-upload metadata (name, size, type — never contents), domain-only matches for shadow-AI discovery, and — for connected Microsoft 365 tenants — Copilot activity metadata (user, host, files read and their labels) plus prompt text for Copilot-licensed users. Never collected: browsing history, other tabs, keystrokes, screenshots, passwords, camera, microphone, location, or Copilot’s responses. Social Security numbers, payment cards, and API keys are replaced with redaction tokens on the endpoint before transmission (and, for Copilot prompts, on our servers before storage) — the raw values are never stored, and captured text purges automatically on your retention schedule (7–365 days).
Yes, in a specific and bounded way: CircleScope produces the AI usage inventory and continuous monitoring evidence that NIST AI RMF assumes an organization already has. The strongest mappings are MAP 1.1 (establishing AI use in context — shadow AI discovery produces the inventory most organizations lack), MANAGE 4.3 (a documented incident-response record for AI misuse: alerts, summaries, resolution notes, audit trail), MEASURE 3.1 (tracking identified risks over time through trend reporting), and GOVERN 4.3 (incidents identified and routed to a named human within about a minute), with supporting coverage for privacy risk examination (MEASURE 2.10) and post-deployment monitoring (MEASURE 4.1). To be clear about the boundary: CircleScope is a monitoring control that feeds an AI governance program — it is not the program itself, and NIST AI RMF alignment is not certifiable by anyone. The full subcategory mapping, boundaries included, is published at circlescope.ai/site/nist.
Yes — CircleScope is purpose-built for MSPs. The Partner Program includes discounted licensing, full white-label (your brand on the portal, alerts, and client reports), PSA integration (Autotask, ConnectWise, HaloPSA) for tickets and metered billing, and deployment through the RMM you already run. One platform serves your whole client base with per-customer isolation. Partners are admitted after a short review before deploying.
One PowerShell script — via RMM, Intune, Group Policy, or manual — sets the managed browser policy, and devices report into the portal within minutes. Deploy to ten computers in about fifteen minutes and you will have real answers by tomorrow. Alerts reach your inbox (or your PSA ticket queue) in under a minute from capture.