VIGIL: An Extensible System for Real-Time Detection and Mitigation of Cognitive Bias Triggers
Researchers' new tool uses LLMs to flag and neutrally rephrase manipulative language as you browse.
A team of researchers led by Bo Kang has introduced VIGIL (VIrtual GuardIan angeL), a novel open-source browser extension designed to combat a subtle form of online manipulation. While tools exist to check facts and source reliability, VIGIL addresses the exploitation of human cognitive biases—like appeals to emotion or false urgency—within text. It performs real-time, scroll-synced analysis on web pages, highlighting potential bias triggers for the user. Crucially, it offers an LLM-powered 'reformulation' feature that can rewrite flagged text into a more neutral tone, with all changes being fully reversible to preserve the original content.
The system is built for extensibility, allowing third-party plugins, and includes several already validated against NLP benchmarks. It offers flexible privacy options, running inference from fully offline on a user's device to cloud-based processing. This positions VIGIL not just as a research prototype but as a practical tool for enhancing media literacy. By making the architecture of persuasion visible and alterable, it empowers users to engage with online information more critically, moving beyond fact-checking to address the underlying rhetorical techniques that can shape discourse.
- First tool to detect & mitigate cognitive bias triggers (e.g., emotional appeals, false dilemmas) in real-time browsing.
- Offers LLM-powered neutral reformulation of text with full reversibility and privacy-tiered (offline/cloud) inference.
- Open-source, extensible system with validated plugins, targeting a gap beyond traditional fact-checking tools.
Why It Matters
Shifts the defense against AI-powered persuasion from just verifying facts to understanding and neutralizing manipulative language patterns.