Inspiration
As an aspiring cyber security analyst, I’ve noticed something worrying: people don’t think twice about what they paste into AI chatbots this inludes things like emails, card numbers, passwords, confidential work details. Once you send the message with that information, that data is stored on someone else’s servers. Depending on your settings, it may be used to train future models. And companies have already had real leaks this way. Since people don't really care, this is an easy lightweight application that can warn people and offer a quick solution.
What it does
paste guard checks what the user has copy and pasted into their chrome websites, it then uses multiple layers to check whether it has private information. Including local checks for info that can be checked with pattern detection, and ai for checking context.
How we built it
Frontend: I made an extension that listens for a users paste. This is then passed into 3 levels of privacy detection, the first 2 are similar. Regex and Luhn: Regex stands for regular expression and its used to detect emails specifically. This is done by checking the users paste, word for word, for a string resembling that of an email e.g letters betwen a-z followed by @ symbol followed by another string of characters etc... Luhn is used to check for card details. The string of numbers 4111 1111 1111 1112 shouldn't be detected but 4111 1111 1111 1111 should be. This is checked with an algorithm designed to validate card numbers (something nearly all payment websites use). The third level of privacy detection is AI, this is used to detect things that cant be picked up by hard coded pattern recognition, instead it looks at context to determine whether its
Backend: made using python uses flask framework holds gemini requests in server
Challenges we ran into
AI token usage
Accomplishments that we're proud of
working project
What we learned
how to create a backend how to use api
