Admunio — AI Data Leak Protection and Prevention

Prevents data leaks to AI tools like ChatGPT, Gemini, and Claude. Runs 100% locally—your data never leaves your device.

Admunio — AI Data Leak Prevention Protect your business from accidental data leaks into ChatGPT, Google Gemini, and Claude. All analysis runs 100% locally on your device. No data is ever transmitted to any server — not even ours. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ WHY YOU NEED THIS ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Your employees are using AI tools every day. And most of the time, no one is watching what they share. Studies show that 1 in 3 employees has pasted sensitive company information into a public AI tool. That includes passwords, API keys, client contracts, internal financial reports, customer data, and confidential business strategy. The problem is not bad intentions — it is habit. People paste information into AI tools the same way they Google something. They do not stop to think about where that data is going. Admunio stops the leak before it happens. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ WHAT ADMUNIO DOES ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Admunio is a Chrome extension that runs silently in the background while your employees use ChatGPT, Google Gemini, and Claude. When someone types or pastes something into one of these AI tools, Admunio checks the content before it is sent. If sensitive content is detected, one of two things happens: → A warning appears, giving the employee a chance to review what they are about to share and decide whether to continue. → The message is blocked entirely, preventing highly sensitive content such as passwords and API keys from being sent at all. If the content is safe, nothing happens. The message sends normally and the employee's workflow is not interrupted. This all happens in under a second, entirely on the employee's device. No data is sent anywhere. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ HOW THE DETECTION WORKS ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Admunio uses two detection layers working together: LAYER 1 — INSTANT PATTERN DETECTION The first layer runs in under a millisecond using pattern rules. It catches well-known sensitive data formats instantly without needing an internet connection or AI model: • API keys (OpenAI, AWS, and other common formats) • Private keys (RSA, EC, SSH, and others) • Passwords and credentials in common formats • Payment card numbers • Email addresses LAYER 2 — LOCAL AI MODEL The second layer uses a real AI model that runs entirely inside your browser using WebAssembly. This model understands context, which means it catches sensitive content that patterns alone cannot identify: • Confidential business strategy and internal plans • Customer names, contacts, and account details • Financial reports and pricing information • Legal and contractual content • Personal data and private employee information • Proprietary technical information The AI model is nli-deberta-v3-small, an open-source model published by Xenova on Hugging Face under the MIT licence. It is downloaded once on first install (approximately 23MB) and cached locally. After the initial download, the model runs fully offline — no internet connection is required for it to work. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ THREE POSSIBLE OUTCOMES ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ✅ ALLOWED — The prompt is safe. Nothing happens. The message sends normally. No interruption. ⚠️ WARNING — The prompt may contain sensitive content. A clear modal appears inside the page, explaining what was found and why. The employee can choose to send anyway or cancel. This outcome is for content that is potentially sensitive but where the employee may have a legitimate reason to continue. 🚫 BLOCKED — The prompt contains clearly sensitive content such as an API key or password. The message cannot be sent. The employee must remove the sensitive content before continuing. Every outcome is shown clearly inside the page. No separate app, no dashboard to check, no notifications. The employee sees the result immediately where they are working. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ PRIVACY — THE MOST IMPORTANT PART ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Most security tools work by sending your data to their servers for analysis. Admunio works differently. Everything happens locally, on the employee's device, inside the browser. ✓ Your prompt text is never sent to Admunio's servers ✓ Your prompt text is never sent to Hugging Face or any other third party ✓ No analytics or usage data is collected ✓ No employee activity is logged or tracked ✓ No personal data leaves the device ✓ The AI model runs offline after the initial download This architecture has two benefits. First, it is genuinely more private — there is no server to breach, no database of prompts to leak. Second, it is faster — local analysis takes less time than a round trip to a server. Admunio's servers never receive your employees' prompt content. There is nothing for us to store, analyse, or accidentally expose. Full details are available in our privacy policy at www.admunio.com/privacy. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ SUPPORTED AI TOOLS ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Admunio currently protects on the three most widely used AI tools in the workplace: • ChatGPT (chatgpt.com) — all conversations including custom GPTs • Google Gemini (gemini.google.com) — the main Gemini chat interface • Claude by Anthropic (claude.ai) — all Claude conversations Admunio only runs on these three domains. It does not run on any other website you visit. This is intentional — we limit the extension's reach to exactly where it is needed. Support for additional AI tools is planned for future versions. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ WHO ADMUNIO IS FOR ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Admunio is designed for businesses and professionals who handle sensitive data and want to use AI tools without the risk of accidental leaks. It is particularly useful for: LAW FIRMS AND LEGAL PROFESSIONALS Lawyers and paralegals often use AI to draft documents, summarise contracts, and research case law. A single accidental paste of a client file or privileged communication can have serious professional and legal consequences. Admunio catches these moments before they happen. ACCOUNTING AND FINANCE TEAMS Finance teams use AI to analyse data, prepare reports, and draft communications. Prompts containing financial figures, client account details, or unreleased results are high-risk. Admunio warns before this content is shared. RECRUITMENT AND HR TEAMS HR professionals handle some of the most sensitive personal data in any organisation — CVs, salary information, disciplinary records, and medical details. Admunio helps ensure this data does not accidentally end up in a public AI tool. DEVELOPERS AND ENGINEERING TEAMS Developers are among the most frequent users of AI tools and also among the most likely to paste API keys, database credentials, and proprietary code into a chat. Admunio blocks the most dangerous of these immediately. MARKETING AND SALES TEAMS Sales and marketing teams work with client names, deal values, competitive intelligence, and internal pricing. Admunio warns when this content is included in AI prompts. SMALL AND MEDIUM BUSINESSES Enterprise companies have IT departments and sophisticated DLP tools. Smaller businesses often have nothing. Admunio provides real protection without requiring a security team, an IT infrastructure, or a per-seat budget. It installs in seconds and works immediately. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ GDPR AND DATA COMPLIANCE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Because Admunio processes all data locally and transmits nothing externally, it is fully compatible with GDPR and other data protection regulations. There is no personal data for us to hold, process, or delete. Employees' prompt content never reaches our infrastructure. This means Admunio does not create the kind of data processing obligations that server-based security tools typically introduce. For organisations that require documentation of data flows for compliance purposes, please contact us at office@admunio.com. Note: Using Admunio does not make your organisation automatically compliant with any regulation. It is a technical safeguard and should be used as part of a broader data protection approach. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ PERMISSIONS EXPLAINED ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Admunio requests only the permissions it needs to function. Here is exactly what each permission is used for and nothing else: STORAGE Used to save your settings locally — whether protection is on or off, and whether strict mode is enabled. These settings never leave your device. UNLIMITED STORAGE Required to cache the AI model (approximately 23MB) in your browser's local storage after it is downloaded. Without this permission, the model would need to be re-downloaded every time you restart the browser. OFFSCREEN Required to run the AI model in a stable background environment inside the browser. The model runs in a hidden offscreen document which provides the consistent JavaScript environment needed for WebAssembly execution. This is a standard Chrome extension mechanism for background processing. HOST PERMISSIONS (chatgpt.com, gemini.google.com, claude.ai) Required to inject the content script that monitors the prompt input field and intercepts submissions on these AI websites. Admunio only has access to these three specific domains — it cannot access any other website you visit. Admunio does not request tabs, browsing history, cookies, bookmarks, or any other sensitive browser permissions. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ HOW TO INSTALL ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1. Click "Add to Chrome" 2. Confirm the installation when prompted 3. Open ChatGPT, Gemini, or Claude 4. Admunio will download the AI model in the background on first visit (approximately 23MB, takes 15 to 30 seconds depending on your connection) 5. Once the model is ready, full protection is active You will see the model status in the Admunio popup — click the extension icon in your Chrome toolbar to check. Once it shows "AI model ready", both detection layers are active. No account required. No email address. No sign-up. Just install and it works. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ OPEN SOURCE AI — FULL TRANSPARENCY ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ We believe in being fully transparent about the technology inside Admunio. The AI model used for classification is nli-deberta-v3-small, published by Xenova on Hugging Face. It is released under the MIT licence, which means it is free to use, inspect, and verify. You can view the model at huggingface.co/Xenova/nli-deberta-v3-small. The model runs using Transformers.js, an open-source library by Hugging Face for running AI models in the browser (Apache 2.0 licence), and ONNX Runtime Web, Microsoft's WebAssembly runtime for AI inference (MIT licence). None of these components send data to external services during normal use. The one-time download from Hugging Face retrieves the model files only — no prompt content is ever transmitted. We are not affiliated with or endorsed by Hugging Face, OpenAI, Google, or Anthropic. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ IMPORTANT LIMITATIONS ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ We want to be honest about what Admunio can and cannot do. Admunio is a risk-reduction tool. It significantly reduces the chance of accidental data leaks but does not eliminate the risk entirely. • The AI model may not detect every instance of sensitive content. Novel phrasing, unusual formats, or content types not represented in the model's training data may not trigger a warning. • Admunio only works on the three supported websites. It does not protect against leaks via mobile apps, desktop applications, API usage, or other AI websites. • An employee who intentionally wants to share data can disable the extension. Admunio is designed to catch accidental leaks, not to prevent deliberate misuse. • Admunio is not a replacement for employee training, a data governance policy, or professional compliance advice. We include these limitations because we believe our users deserve an accurate picture of what they are installing. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ FREE TO USE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Admunio is currently free for both personal and business use. Visit www.admunio.com for more information, our privacy policy, and our full terms of service. Questions or feedback: office@admunio.com