AI Browser Guard - Privacy Policy

Last updated: June 6, 2026

1. What this extension does

AI Browser Guard is a Chrome extension that detects and monitors AI agents operating in your browser sessions and blocks their scripted navigations, form submissions, and downloads best-effort for in-page automation. It provides:

• Detection of automated browser control (WebDriver flags, CDP connections, behavioral analysis)
• A kill switch that closes the tabs an agent controls
• A delegation wizard to define access boundaries for AI agents
• Boundary violation alerts when agents attempt unauthorized actions
• A session timeline logging agent activity

2. Data we process locally

We do not collect any personal data.

The extension processes the following information entirely on your local device. By default none of it leaves your browser:

• Detection signals: Browser automation indicators such as navigator.webdriver flags, CDP markers, and event timing patterns. These are analyzed locally and never transmitted.
• Session logs: A chronological record of detected agent actions (timestamps, action types, target URLs, CSS selectors, and outcomes). The last 5 sessions are stored locally.
• Delegation rules: Access control rules you configure (presets, site patterns, time bounds). Stored locally in chrome.storage.local.
• Violation records: Logs of boundary violations (what was attempted, which rule blocked it). Stored locally, limited to the 100 most recent entries.

2a. Optional network features (off by default)

The extension includes four optional features that make network requests. All four are disabled by default. A fresh install makes zero network requests. Each feature only acts after you explicitly enable it, and you can turn it off again at any time with one click.

Three of them talk to OpenA2A servers. The fourth, site safety declarations, is different: it requests a file from the website your agent is on. It is described in full below, and it is the only feature that contacts a server OpenA2A does not operate.

• AIM identity lookup (setting aimLookupEnabled, default off): when an agent is detected, the extension can query aim.opena2a.org for a published identity and trust score for that agent type. The request contains only the detected agent type string (for example playwright). The page origin is used only as a local cache key and is never sent.
• Registry trust lookup (setting registryLookupEnabled, default off): the extension can query api.oa2a.org for a registry trust score for the detected agent type. The request contains only the detected agent type string.
• Anonymized contribution (consent default off, with a one-time tip after 5 detections that you can dismiss): if you opt in, the extension can send anonymized detection and behavior summaries to api.oa2a.org to improve community threat intelligence. Each contribution contains an anonymous contributor token (a per-install random identifier generated once and stored locally, for deduplication only, that does not identify you), the detected framework name, detection summaries, behavior summaries (interaction count, success rate, anomaly count), the extension version, and timestamps.
• Site safety declarations (setting aiSafetyTxtEnabled, default off): if you opt in, then when an AI agent is detected on a page, the extension requests one file from that website: /.well-known/ai-safety.txt. This is a public file a site may publish to state what it claims about its own content, using the ai-safety.txt format. What the extension shows you is what the site claims about itself. It is not verified, and a declaration never changes what the extension detects or blocks.
  • What is sent: nothing beyond the request itself. No cookies, no page address, no referrer, no identifier, and nothing about you. The request is a plain GET for that one fixed filename.
  • Who it reaches: only the site your agent is already on. The extension will not follow a redirect to any other server, and only requests declarations over HTTPS.
  • What the site can infer: because your agent has already loaded a page there, that site's logs already record your visit. The new information the site gains is that something on your device checked for a safety declaration, which it could use to tell visitors running a declaration-reading tool apart from those who are not. This is why the feature is off by default and asks for your consent.
  • Only while an agent is running: nothing is requested for pages you browse yourself. The check happens only when an AI agent has been detected on that page.
  • Caching: a result is stored locally so repeat visits do not repeat the request. That is 24 hours when the site gave a usable answer, and 5 minutes when the site could not be reached, so a site that is temporarily down is retried sooner. Storage is capped at 50 sites. Turning the setting off deletes every stored declaration.

These features never transmit your URLs, page content, form data, keystrokes, cookies, authentication tokens, personal identity, or browsing history. To enable or disable them, use the extension settings. Disabling a feature stops all outbound requests for it immediately; disabling contribution also clears any queued events.

3. How data is stored

All data is stored in chrome.storage.local, which is a browser-provided storage mechanism that keeps data on your device only. Data is not synced across devices.

• Session history: Last 5 sessions retained
• Detection log: Last 100 entries retained
• Delegation rules: Retained until you change or remove them
• User settings: Retained until you change them or uninstall the extension

Uninstalling the extension deletes all stored data.

4. Data we never collect or transmit

Regardless of which optional features you enable, the extension never collects or transmits:

• No personal information (name, email, account credentials)
• No browsing history, full URLs, or page content
• No form data, keystrokes, or user input
• No cookies, authentication tokens, or session identifiers
• No usage analytics, behavioral telemetry, or tracking
• No crash reports or error telemetry

5. Network communication

By default, this extension makes zero network requests. All detection, analysis, and enforcement happens locally in your browser, with no external API calls beyond Chrome's built-in extension update mechanism.

Four optional network features are off by default and act only after you explicitly enable them. Three contact OpenA2A servers: AIM identity lookup and registry trust lookup (each sends only the detected agent type to query a published trust score), and anonymized contribution (sends framework names and numeric detection summaries, never URLs or page content). The fourth, site safety declarations, requests /.well-known/ai-safety.txt from the website an AI agent has been detected on, so with that setting enabled the extension can contact a website you visit, not only OpenA2A servers. It sends no cookies, no page address, and nothing about you, and it will not follow a redirect to any other server. Each can be turned off again with one click. Disabling contribution clears any queued data, and disabling site safety declarations deletes every stored declaration.

6. Permissions explained

The extension requests the following Chrome permissions:

• storage: Store session logs, delegation rules, detection history, and settings locally on your device.
• alarms: Schedule periodic background tasks: delegation rule expiration cleanup, a service-worker keepalive, a detection sweep that re-checks for attached automation, and a flush of the optional contribution queue (which sends data only if you have enabled that feature).
• notifications: Display boundary violation alerts when an agent attempts an action outside your delegation rules.
• debugger: Detect when an automation framework (Playwright, Puppeteer) attaches a Chrome DevTools Protocol debugger, a core detection signal, and, for tabs you have delegated to an agent with restricted boundaries, attach a CDP session to block network requests that violate those boundaries. The extension does not use the debugger to read page content or inject code, and detaches when delegation ends.
• downloads: Monitor downloads created while an AI agent is active so agent-initiated downloads can be detected and cancelled per your delegation rules (your own downloads are never flagged), and save a session report as a JSON file when you choose to export it.
• host_permissions (<all_urls>): The content script must run on all pages to detect automation frameworks and enforce delegation boundaries, since AI agents can operate on any website. Detection cannot be limited to specific domains without losing coverage. Page content is analyzed locally and is never transmitted.

7. Third-party services

This extension does not use or integrate with any third-party analytics platforms, advertising networks, or trackers, and it sends no data about you to any third party.

The OpenA2A servers it can contact, and only when you opt into the features in section 2a, are aim.opena2a.org and api.oa2a.org.

If you enable site safety declarations (section 2a), the extension additionally requests one public file, /.well-known/ai-safety.txt, from a website an AI agent has been detected on. That website is not a third-party service the extension integrates with, and it receives nothing about you, but it is a server OpenA2A does not operate and it can see that the request was made. This is stated here rather than omitted because an earlier version of this policy said OpenA2A's own servers were the only ones the extension could ever contact. With this setting enabled, that is no longer accurate.

8. Children's privacy

This extension does not knowingly collect any information from anyone, including children under 13. Since no personal data is collected, COPPA considerations do not apply.

9. Changes to this policy

If this privacy policy changes, the updated version will be published at this URL with a new "Last updated" date. Material changes affecting data collection would require a new extension version review by the Chrome Web Store.

10. Contact

For questions about this privacy policy or the extension, contact:

• GitHub: github.com/opena2a-org/AI-BrowserGuard/issues
• Organization: OpenA2A