SafisAI

Detects and indicates AI usage on websites, providing transparency about AI-generated content and active AI agents.

SafisAI serves as a local visibility layer for the modern AI web, prioritizing user transparency without compromising privacy. Every scan is performed entirely on-device; the extension operates without a backend or remote servers, ensuring that your page content, prompts, and browsing history never leave your browser. By analyzing the technical structure and behavioral patterns of a page it identifies AI presence based on real-time, local evidence. The Detection Engine The detection engine categorizes findings into four distinct areas to provide a comprehensive view of AI involvement. It identifies active AI Workflows, such as chat-style assistants or generative tools, by scoring UI patterns and technical fingerprints even if the specific platform is unknown. Beyond active tools, the extension surfaces broader AI Presence through script detection and chatbot widgets. Additionally, the AI Image analyzer probes media for provenance markers and C2PA metadata to help verify if an image was generated or modified by AI. Evidence and Privacy To maintain absolute credibility, SafisAI utilizes a strict evidence model that labels every finding as Observed (direct technical detection), Disclosed (site-reported text), or Possible (logical inference). This approach prevents alarmism by showing you not just if AI is present, but exactly how it was found. A major priority of the tool is surfacing Data and Privacy signals, such as active session recorders or analytics, specifically when they are used alongside AI agents where prompt leakage might occur. Implementation and Boundaries Technically, SafisAI runs lightweight, snapshot-based scans that are capable of penetrating the Shadow DOM to find hidden agents that standard detectors often miss. The system is designed for efficiency, automatically skipping hidden tabs and clearing its local cache upon browser restart to keep your session clean. It is important to note that SafisAI is purely an informational tool: it does not block content, alter page behavior, or make decisions on your behalf. It simply surfaces the facts, leaving the final decision of how to interact with the AI web entirely to you. Heads up, this is early development. Whether you stumbled across SafisAI or someone passed it your way, please remember it's a v1.0.0 — the first build we're putting in front of real testers. The AI web is a moving target: new agents launch every week, sites shift their UI overnight, generative content gets re-hosted in places no one labels. Detection on something this slippery is genuinely hard, and SafisAI will sometimes miss things, sometimes flag things it shouldn't, and occasionally do both on the same page. We'd rather be honest about that than pretend we've solved it. If you spot a misfire, a missed AI surface, a bug, or just a weird wording choice, please tell us at safisai.com/report-bug. Every report is read, and the detection gets a little tighter with each one. Thanks for helping us figure this out.