Inspiration
With government censorship and monitoring becoming an active issue across the world, finding methods for secure communication is paramount. With prevalent methods such as RSA, one can easily tell when a message is encrypted, which can lead to users and communities being flagged. This hinders important activist movements and expressions of free speech, which are basic rights and critical to the advancement of our society. Our project, which utilizes Steganography, offers encryption that evades clear encryption flags by mimicking natural speech, therefore protecting the privacy of users. Cipher AI is an AI assisted steganography tool engineered to create better concealed encryptions at ease.
What it does
The user inputs a secret message, which is then communicated to the backend. This backend takes a database of tweets and computes a list of viable key words based on word frequency. This is then mapped to the user-input secret message, which is then placed throughout a generated text to be returned to the user. Thus, the user receives a sentence or paragraph of natural-sounding text, which is the encrypted message, and a password-protected JSON file that contains the decryption key.
How we built it
Our software runs through a React-based frontend with Python supporting our backend. We also utilize a large subset of the Sentiment140 dataset, sourced from Stanford, to help with optimizing our key generation. We then prompted OpenAI's GPT-4o to create a dictionary and encoded message.
Challenges we ran into
Some of the primary challenges we ran into while working on this project were formatting and filtering the data, getting our key generation to work properly, and linking our frontend and backend work together.
Accomplishments that we're proud of
We are proud of trying to tackle an open research problem! While linguistic steganography is widely used in casual, verbal contexts (like magicians communicating with their assistants), there are very little, if any, existing models that allow users to create simple covert codes that can be easily used in the platforms we utilize throughout our daily lives, like social media. We were able to build an end-to-end application, which for some of our team was our first experience with full-stack development.
What we learned
As a team, we learned a lot while working on this project. From researching the theory of linguistic steganography its implementations, to learning React and linking our front end, almost all aspects of going about building this project provided an exciting challenge for our group. Thankfully, we also gained another key skill through this process - the ability to harness each group member’s strengths to work through individaul obstacles as a team.
What's next for Cipher AI
As of now, our software is just a piece of the puzzle for providing covert ways of secure communication. Implementing this into a fully functional program would include working through other communication platforms to send the encoded messages. Additionally, improving our dataset and integrating multiple layers of encryption would allow us to create a more realistic and secure application. In order to validate the undetectable nature of the code generated by GPT-4, we would also like to implement another AI model in tandem with GPT-4 to reduce possible bias within AI models. Overall, we envision a future of Cipher AI in helping oppressed communities work together to challenge boundaries and jumpstart the growth of our society.
Log in or sign up for Devpost to join the conversation.