The Governance 1st Browser Extension sits on top of every public AI tool your employees actually use, ChatGPT, Claude, Gemini, Copilot, Perplexity, DeepSeek, Meta AI, Grok, watching prompts before they're sent and responses before they're trusted. Ten bad behaviors, 200 underlying guardrails, one bridge into Governance 1st when you're ready to operationalize.
Most "AI monitoring" lives behind the API. Useful for the AI you procured. Blind to the AI your employees pasted a customer record into at 11pm. The extension closes that gap by living inside the browser, on the same screen the AI is being used on, watching what gets typed and what gets returned.
No server to stand up, no agent to deploy, no SDK to integrate. Install once and it's live on every supported AI tool the next time you open one. Six concrete capabilities, all configurable from the popup.
Pattern library catches sensitive data before it leaves the browser. Credit cards, SSNs, source-code secrets, internal URLs, the categories you configure. Soft-warn or hard-block, your call.
Counts visits, prompts, flagged items, redacted items, and behavior flags per session, per day, per LLM host. The number your auditor wants and your CISO doesn't have today.
From a library of 200 guardrails, the extension picks the one that applies to what the user just typed and shows it to them right there, next to their prompt.
Routes the LLM's response through an independent judge model that scores it against the 10 behavior categories. Hallucination, drift, bias, dangerous advice, jailbreak success, all ten. Logged with the prompt.
Optional enforcement mode. When a filter or behavior check fires, the prompt is held and the user gets a one-screen explanation of which rule triggered and why. No more "I didn't know I shouldn't."
Enterprise tier saves the same activity records to your organization's Governance 1st workspace. If you already run a governance program there, the data lands in your existing structures. If you don't, this is your starting point, the first real record of how AI is being used and a foundation to build the rest of governance on top of.
The behavior library is HR Rebooted's patent-pending taxonomy of how LLMs go wrong in production. Ten categories, 200 underlying guardrails. The same taxonomy Governance 1st uses for output evaluation, so what your browser sees and what your platform sees speak the same language.
The model invents facts. Cites papers that don't exist, names people who weren't there, fabricates statistics, quotes that were never said. 20 rules covering grounding, citation enforcement, math audits, temporal validity, and entity coherence.
The same model gives meaningfully different answers to the same prompt over time. Silent regression after a vendor update. 20 rules covering semantic baseline tracking, format conformity, perplexity audits, and shadow-deployment regression.
Systematic asymmetry in how the model treats people based on demographic attributes. 20 rules covering demographic parity, occupational stereotyping, geographic decentering, name-based scoring, and counterfactual substitution.
Confident answers in domains where confident wrong answers hurt people. Unlicensed medical diagnoses, financial picks, legal strategy, CBRN, malware, lethal-force optimization. 20 rules with hard blocks across these categories.
Prompts engineered to break the model out of its guardrails. Direct, indirect, roleplay-laundered, Base64-encoded, multilingual pivots. 20 rules covering system prompt encapsulation, suffix filtering, DAN pattern matching, and dual-LLM separation.
Over-refuses safe requests, under-refuses unsafe ones. Lectures the user when it shouldn't, complies when it shouldn't. 20 rules covering false-positive override, intent classification, tone audit, and granular reason labeling.
Slurs, dehumanizing analogies, genocide rationalization, eugenics, transphobic rhetoric, dogwhistles. The hard-floor stuff that can never reach a customer screen. 20 rules with near-zero tolerance toxicity classifiers.
Confidently delivered but structurally broken reasoning. Ad hominem, straw man, slippery slope, false dichotomy, correlation-as-causation. 20 rules to catch the ones that read fluent but fall apart on inspection.
The model picks a political side. Candidate endorsements, loaded adjectives, asymmetric treatment of left-vs-right policy. 20 rules covering multi-perspective balance, source disclosure, and partisan anchor detection.
Confident assertions that contradict verified consensus. Anti-vax medical claims, election fraud narratives, climate denial, fabricated quotes, predatory MLM scripts. 20 rules with consensus alignment and integrity shields.
The extension works automatically on the AI tools your employees are most likely to use. The moment someone opens one in their browser, it's watching. As new AI tools become popular, we add them.
The Governance 1st Browser Extension is built on patent-pending HR Rebooted technology, from the 10-behavior, 200-rule guardrail library to the independent AI agents that score every prompt and response in real time. It all runs in your browser. Nothing to install on your computer beyond the extension itself. On the individual tier nothing about your activity ever leaves your device. On the enterprise tier, activity records are saved to your organization's Governance 1st workspace where they become the foundation of a governance program, whether one already exists or you're building it from here.
Built on Manifest v3 and supported across Chromium-based browsers (Chrome, Edge, Brave, Arc, Opera). Installs from your browser's extension store or sideloaded by IT. No driver, no kernel hook, no special privileges.
Scripts run scoped to the LLM domain. They read the DOM the user already sees and modify the prompt before submission. Same trust boundary as the page.
Filter library, guardrail library, and activity logs live in the browser's local storage by default. No external server required for the individual tier to work end-to-end.
Behavior monitoring routes every prompt and response through separate AI agents that score it against the 10 categories. Independence is the point. The AI being judged is not the AI doing the judging.
The 10-behavior, 200-rule guardrail library and the judging architecture that runs on top of it are HR Rebooted intellectual property. The same technology powers the evaluator suite inside Governance 1st.
Single outbound channel to api.hrrebooted.com when a Governance 1st workspace is configured. Activity records, not prompts. The data is stored in your workspace, ready to plug into existing governance structures or to seed new ones.
Most enterprises know employees use public AI tools. None of them have the data. The extension is the cheapest, fastest way to start capturing it, and the workspace where that data lands (Governance 1st) is the one your committee, your CISO, and your auditor will eventually need anyway.
The same browser extension, deployed across your workforce. Every prompt and response is saved to your organization's Governance 1st workspace, where the activity becomes the foundation of a real governance program: an inventory of which AI tools are actually being used, the first evidence trail an auditor can read, and the data behind any dashboards or controls you choose to add on top. Whether you already run governance or you're starting now, the 10-behavior taxonomy speaks the same language at both ends.
Individual users install the extension from their browser's extension store or sideload the latest release. Enterprises talk to us about a centrally managed deployment and integration with the Governance 1st platform.