Preempt AI - AI is powerful. Make sure it's safe too.

Hey Product Hunt! 👋 I am launching Preempt AI v2 today — a security layer for AI interactions that works for both everyday users and developers building on LLMs. --- The problem AI assistants are powerful. But they're wide open to abuse: - People accidentally paste SSNs, credit cards, and patient data into chat boxes - Attackers embed hidden instructions in documents to hijack AI behavior - Jailbreak attempts bypass safety measures — exposing users and developers to risk Most people have no idea this is happening. --- What Preempt does Scan every prompt before it leaves your keyboard. For users → install the browser extension. A "P" icon appears in every AI chat input and silently scans in real time. Nothing changes about how you use AI — you just get warned before something goes wrong. For developers → one call to /v1/sanitize. If a prompt is malicious, it's blocked. If it contains PII, it's encrypted — and real values are restored in the response. Zero infrastructure changes. --- What's new in v2 🧠 Smarter ML — trained on 16,000+ real in-the-wild jailbreak prompts + 589,000+ prompt injection samples from a security research dataset. Now catches evasion tricks: typo substitution, Unicode obfuscation, roleplay framing, gamification, and model impersonation. 🌐 Multi-browser — live on Chrome and Microsoft Edge. Firefox coming next. 📄 Document scanning — upload PDFs, DOCX, images, or CSVs before sharing with AI. Extracts text, scans for hidden injections, flags PII. Yes — people really do hide prompt injections inside documents. 📊 99.65% ML accuracy · <10ms latency · 41+ attack types · 12+ languages --- Try it 🧩 Chrome Extension → https://chromewebstore.google.co... 🧩 Microsoft Edge Extension → https://microsoftedge.microsoft.... 🔌 Live Demo - https://www.getpreempt.com What attack type worries you most when using AI? Happy to discuss below 👇