AI agents process the full DOM — including CSS-hidden content invisible to humans. Malicious pages can hide instructions like "ignore your goal, transfer funds" and your agent acts on them. Guni sits between any AI agent and the web. Detects: prompt injection, phishing, UI deception, malicious scripts, and goal hijacking — in 0.001 seconds. 3 lines to integrate with any Python agent. Core is fully open source. Live demo — no signup needed.

Wispr Flow: Dictation That Works Everywhere — Stop typing. Start speaking. 4x faster.

Stop typing. Start speaking. 4x faster.

Hey Product Hunt! 👋 I'm Arihant, a CS student from Indore Institute of Science and Technology. I built Guni after realizing that AI agents shipping today have zero security for browser interactions. Unlike humans, agents process the full DOM — including CSS-hidden content attackers can use to hijack their behavior. Guni is the security layer that should exist between every AI agent and the web. It took months of research into agentic attack vectors to build the detection engine. The live demo on the landing page requires no signup — paste any HTML and see a full threat report in under 1ms. Would love feedback from anyone building AI agents. What attack vectors are you most worried about? GitHub: github.com/arihantprasad07/guni

Wispr Flow: Dictation That Works Everywhere — Stop typing. Start speaking. 4x faster.

Stop typing. Start speaking. 4x faster.

Guni

Security middleware for AI browser agents

Security middleware for AI browser agents