Don't Feed the AI (github.com)

A new tool called "DontFeedTheAI" has been developed to enhance data privacy when utilizing AI language models. This proxy solution ensures that sensitive information, such as IP addresses, credentials, and personally identifiable information (PII), is stripped from every request sent to AI services. It restores this information upon receiving responses, allowing users to interact with AI tools without risking exposure of their real data. This is particularly significant for pentesters, developers, legal consultants, and researchers who require AI insights while keeping their client or internal data confidential. The architecture of DontFeedTheAI operates on a dual-layer system, with local detection capabilities that prevent sensitive information from traversing beyond the user’s machine. Unlike traditional cloud anonymization APIs that expose data to third parties, or local models that are oblivious to sensitive details, this tool combines the best of both worlds by utilizing cloud AI without compromising data integrity or security. By offering detailed auditing and automated improvement features, it empowers users to validate coverage and tighten security measures efficiently, making it a valuable asset for those in regulated or security-sensitive environments.