ChatGPT web · managed browserSynthetic input
Protect the customer detail, keep the request useful.
Summarise the renewal notes for Morgan Hale. Contact: morgan.hale@example.com. Keep the next steps concise.Download prompt fixtures
AI SECURITY COVERAGE
AI security coverage
Know where you stand.
See where Aona discovers AI use, applies prompt policies and protects files. Find the right coverage for your employees’ browsers, desktop apps and AI agents.
For security and IT teams planning a workforce AI rollout.
Choose your employee surface
Chrome, Edge, Firefox & Safari
Inside the browser
Production pluginDiscover AI use
On devices with the plugin installed
Apply prompt policies
On supported AI submission paths
Redact sensitive files
DOCX, XLSX and PDF upload paths
Start with ChatGPT, Claude, Gemini, Copilot or Bing. Confirm the exact interface, action and policy, not just the tool name.
Coverage guide · not a live device scan
FOUR DIFFERENT QUESTIONS
A known tool isn’t
a protected workflow.
Separate catalogue recognition, observed use and tested policy outcomes. Each needs its own evidence.
- 01
Known
The tool is in the catalogue.
Not proof of employee use. - 02
Observed
Activity appears on a deployed endpoint path.
Not proof of a policy action. - 03
Evaluated
A configured policy evaluates a prompt or file.
Validate the exact route. - 04
Verified
A current test documents the outcome.
Keep its scope and evidence date.
THE COVERAGE REFERENCE
Find your path.
Check the details.
Start with the capability you need. Then validate the app, endpoint and action that make it work.
AI tool discovery
10,000+ AI tools tracked
- Catalogue breadth
- Confirm observed use through the installed browser plugin or native app. No endpoint means no Shadow AI visibility.
Browser prompt policies
Chrome · Edge · Firefox · Safari
- Supported routes
- Active governance covers top-tier ChatGPT, Claude, Gemini, Copilot and Bing. Test the exact interface, policy and submission action.
File redaction
DOCX · XLSX · PDF
- Documented formats
- Check the upload route, sensitive-data replacements and resulting document layout. Format support is not every possible file.
Native AI applications
Windows & macOS
- Release-specific
- IT teams can deploy the Aona endpoint app on Windows and macOS with their preferred software deployment tools. Intune is one option; direct installation is also available on both platforms. Confirm the provider, OS, input path and installed release for the controls you need.
AI agent inspection
Native endpoint app
- Limited rollout
- Confirm runtime and privacy scope. Inspection does not imply remote agent blocking, fleet-wide scanning or agentless API coverage.
Unverified stays unverified. Record the release, configuration and evidence date before expanding a rollout.
DATED PRODUCT EVIDENCE
See what was checked.
These checks used a built Chrome plugin on simulated AI pages and the Aona file redaction engine on local synthetic documents. They did not use a live AI provider, an installed customer endpoint or the production policy API.
EVIDENCE RECEIPT
Browser fixtures and three local file formats
APP, CLIENT AND ACTION
A narrow coverage cohort
A passing mock test describes its fixture. It does not stand in for a live provider or your installed client.
| App or source | Client | Action checked | Status | Limit |
|---|---|---|---|---|
| ChatGPT web fixture | Built Chrome plugin | Prompt rewrite, clean pass and file drop block | Passed 25 Sep 2026 | Simulated provider page and scripted policy replies. No live ChatGPT or production policy decision was tested. |
| Claude web fixture | Built Chrome plugin | Completion interception and scripted block | Passed 25 Sep 2026 | Simulated page and policy reply. No live Claude account or deployed client was tested. |
| Copilot web fixtures | Built Chrome plugin | Consumer WebSocket block and Microsoft 365 SignalR block | Passed 25 Sep 2026 | Two simulated Microsoft surfaces. Other Copilot clients, accounts and file paths need separate tests. |
| Local synthetic files | Orbital Files Edge engine, offline | Seven supplied replacements in DOCX, XLSX and PDF | Passed 27 Sep 2026 | No sensitive-data detection, policy selection, browser upload or production API was exercised. |
The 25 September Chrome smoke was CI run 36080209260 using built plugin revision 0bd184e, simulated provider pages and scripted policy replies. The 27 September offline file check used engine revision 013b7e9; its focused suite passed 19 of 19 tests. Full revisions, file hashes and checks are recorded in the manifest. Inspect the file manifest
SYNTHETIC BEFORE AND AFTER
Inspect the saved files.
Each pair is a fictional customer-service record. A local Aona engine used seven supplied replacement values. The original and resulting native files are available without a form.
- 01
DOCX
Seven supplied identifiers were replaced. The saved Word file opens with its headings, table and order reference in place.
- 02
XLSX
Seven supplied identifiers were replaced in the saved workbook. This fixture uses a plain colon in the title.
- 03
PDF
Seven expected sensitive spans are visibly masked in the checked image-only output. The page layout and order reference remain.
What this sample does not prove
- The file pairs show caller-provided replacement values in the offline engine. They do not demonstrate detection, a configured policy, a browser upload, a production API response or identical results for another document.
- The tested PDF output is image-only. It has no selectable or searchable text layer; its visual layout remains readable in this one-page example.
- A separate offline XLSX check changed an em dash into visible numeric character-code text in a title. The published workbook uses a plain colon and does not establish general character fidelity. Review exact characters in your own test.
- This cohort does not test live Gemini or Bing actions, Firefox or Safari workflows, or an installed Windows or macOS native release. Product scope and client behavior must be checked for the intended route.
- The Chrome CI summary marked one panel-wrapper check flaky and skipped live storage parity. The passing simulated-page checks do not settle those paths.
SAMPLE SECURITY REPORT
A readable review with a defined denominator.
The fictional 30-day review counts 20 evaluated policy outcomes across Finance and Operations: 12 compliant, four redacted, three blocked and one warned. Its denominator is 20 evaluated outcomes, not employees, files or all AI activity.
The report and ledger are invented examples, not Aona exports or customer results. NO_POLICY is outside this denominator, and the public events API does not include compliant records.
EMPLOYEE AI SECURITY REVIEWFICTIONAL
11 Aug - 9 Sep 2026
20evaluated sample outcomes- Compliant
- 12
- Redacted
- 4
- Blocked
- 3
- Warned
- 1
EVALUATE YOUR PATH
Bring one app and action to a focused demo.
We can map the installed client, policy response, saved output and evidence your team needs to verify.
Choose the control you actually need.
Service access, account access and sensitive-data protection answer different questions. Test the one your policy depends on.
- 01Block the service
- Can this device reach the AI service?
A domain, category or application rule can deny access on the network or client paths it controls.
It does not distinguish personal and work accounts on the same service, or inspect the content of a permitted prompt.
- 02Restrict the account
- Can employees use only the approved workspace?
An account or tenant restriction needs a supported identity, browser or security-product feature for that specific service.
Verify allowed and denied accounts, sign-in changes and bypass paths. A domain block alone is not an account restriction.
- 03Inspect the content
- Can this prompt or file be sent?
A supported DLP path evaluates the information against the configured policy and applies its available response.
Test typed text, pasted text and uploads separately. Content inspection does not by itself restrict the signed-in account.
YOUR EVALUATION, MADE PRACTICAL
Start with a test.
Leave with evidence.
Use a synthetic example, agree the result you need, then record what actually happened.
WHAT TO CHECK
- 01
Select the person-name and email rules for the test group.
- 02
Confirm the configured response before the request is sent.
- 03
Compare the employee notice with the recorded policy outcome.
Record the browser, extension version and ChatGPT interface. Repeat typed, pasted and attached inputs separately.
Get the acceptance checklist (PDF)Client contract checks 2026-09-19
These checks run real client functions with mocked dependencies. They do not prove detection accuracy, upload coverage, an installed release or end-to-end Sentinel ingestion.
Policy action handling
Block priority, redaction responses and silent flags with supplied policy responses.
12/12 checks passedChatGPT document request
DOCX restoration request uses the stateless replacement contract.
1/1 checks passedClaude document request
DOCX restoration request uses the stateless replacement contract.
1/1 checks passedMAKE COVERAGE TANGIBLE
Bring one workflow.
Leave with a clearer decision.
A useful coverage review follows the work your people actually do.
- 01
Name the path
AI tool, browser or app, OS, and intended action.
- 02
Test with synthetic data
Run the prompt or upload against the chosen policy.
- 03
Record the decision
Review the outcome, evidence and unresolved gaps.
YOUR COVERAGE REVIEW
One path. A complete picture.
- Application
- ChatGPT
- Employee surface
- Chrome + Aona plugin
- Input to test
- DOCX with synthetic identifiers
- Policy action
- Redact before upload
Inspect the redacted outputReview recorded policy evidenceDocument any unsupported behavior
Illustrative test plan, not a completed test or coverage guarantee.
COVERAGE MEETS DEPLOYMENT
Where it runs
is a separate choice.
Choose backend hosting and prompt processing independently. Verify that your required features support the selected combination.
01 / BACKEND HOSTING
Customer cloud / On-premises / Aona-managed
02 / PROMPT PROCESSING
Aona’s configuration does not determine where a third-party AI provider processes data.User device (edge) / Customer cloud or on-premises / Aona-managed
BEFORE YOUR PILOT
The questions
worth asking.
Does a catalogued AI tool mean it is discovered or controlled?
Aona tracks 10,000+ AI tools in its catalogue. Catalogue recognition is not observed employee use or active enforcement on your devices; those require an installed client and a supported application and input path.
What must be deployed to observe employee AI use?
The applicable browser plugin or native endpoint client must be installed and configured on the managed device. Aona does not provide an agentless network, identity-log, or SaaS-integration scan for this purpose.
Which browser and prompt-policy paths are supported?
The Aona browser plugin supports Chrome, Edge, Firefox and Safari. Confirm the intended browser, application and input path; browser support does not establish feature parity. Active prompt policy enforcement is focused on top-tier ChatGPT, Claude, Gemini, Copilot and Bing paths; confirm the exact interface, action, data category and current release before rollout.
Which file formats can be evaluated?
DOCX, XLSX, and PDF are documented layout-preserving redaction formats. Validate the intended upload route, policy, and resulting file with a synthetic test.
What are the native and agent limits?
Native provider, transport, OS and release coverage varies. IT teams can deploy the Aona endpoint app on Windows and macOS with their preferred software deployment tools. Intune is one option; direct installation is also available on both platforms. AI-agent inspection is native-endpoint only and in limited rollout; it is distinct from agent-action control.
Does backend hosting determine prompt processing?
No. Backend hosting and prompt processing are independent choices. Confirm the supported feature and integration combination for the selected configuration, and assess third-party AI-provider processing separately.
START WITH YOUR ENVIRONMENT
Map coverage for your environment
Bring the app, surface, endpoint, policy action, and file type you need to evaluate. We will map the current evidence and name what remains unverified.