Cookie Crumblr

Main screen when everything is protected!
List of cookies on system.
Main screen when extension is turned off.
Information screen explaining all the settings.

Inspiration

We didn’t start Cookie Crumblr because we wanted to build a browser extension. We started it because we saw the data.

We were researching Real-Time Bidding (RTB) for a CS class and stumbled into FTC enforcement actions, ICCL complaint databases, and EFF research on ad-tech exploitation. The numbers were staggering; Hundreds of billions of data broadcasts per day. But while the abstract loss of privacy is bad enough, that's not our biggest concern. The real-world impact of data exploitation is.

Sold user location data has already been used to track protesters at demonstrations, a fact Google has already had to fight in court.

The harms are systemic:

People in financial crisis are bombarded with predatory products
Health seekers are targeted with exploitative ads
LGBT+ individuals could be outed through purchased location histories
Religious minorities faced discrimination via targeted content
Activists tracked in real-time through ad networks
Racial minorities experiencing algorithmic discrimination in housing, credit, and employment

Many cookie-blocking apps already exist but they are usually complex to understand, and simply block or delete cookies which websites can work around and re-download. This realization inspired us to protect individual users in a way that's simple, poisons already existing profiles built for users, and innovative. Instead of blocking tracking, we asked: What if we could make the data itself useless?

What it does

Cookie Crumblr is a Firefox extension designed to protect users from exploitative tracking by poisoning the data that trackers rely on.

AI-powered cookie classification: identifies functional cookies from deceptive analytics and harmful tracking cookies in real time
Automated scrambling of tracking cookies with random data that preserves size and encoding
Clipboard protection: clears sensitive data after a set amount of time
User-friendly dashboard: shows cookie count, safety status, and simple controls
Continuous learning: new cookies are automatically classified and handled

In practice, users browse normally while Cookie Crumblr corrupts the profiles that trackers try to build, making behavioral targeting ineffective.

How we built it

Research & Problem Validation

Mapped RTB ecosystem and real-world harms
Analyzed data on predatory targeting
Documented case studies: protest tracking, financial exploitation, etc
Defined core users: children, non-technical adults, and vulnerable populations

AI-Powered Classification

Trained a K-Nearest Neighbor (KNN) model on labeled cookie datasets
Features: cookie names, domain ownership, known tracking networks, payload structure
Achieved real-time classification with std_deviation: 0.05866 and a k-value of 11.

Extension Development

Reads cookies in real time via Firefox's native tools
Sends cookies to backend AI for classification
Scrambles harmful cookies while preserving format
Dashboard shows cookie status, user controls, and real-time metrics
Continuous learning pipeline updates model with new cookies

Accessibility & UX

One-click activation, install-and-forget
Color-coded dashboard
Plain-language labels instead of technical jargon
Keyboard navigation and multilingual support
Optional advanced scrambling settings for more technical users

Challenges we ran into

Legitimate analytics vs. harmful tracking: solved by making the AI evaluate analytics cookies on a case-by-case basis
Cross-browser consistency: Firefox MVP first; Chrome, Safari, and mobile might require different implementations for grabbing cookies

Accomplishments that we're proud of

Built a fully functional Firefox extension with real-time AI classification and cookie scrambling
Developed a scalable data-poisoning strategy for cookies
Created a highly accessible, non-technical UX

What we learned

The tracking ecosystem is more complex than we thought, requiring selective cookie handling
Cookies are not labeled consistently across sites, leading to a lot of data cleaning during training and methods to overcome this when classifying
Privacy isn’t just technical: vulnerable populations need trust, literacy, and support
Small UX improvements (plain language, color coding) significantly increase adoption

What's next for Cookie Crumblr

Integrated continuous learning to handle new tracking patterns automatically
Multi-browser support: Chrome, Safari, Edge
More privacy tools for invasive tracking
Community trust: open-source classifier and feedback system
Regulatory advocacy: protect data poisoning as a privacy defense

Our goal: make the tracking system fail for everyone, not just tech-savvy users. When enough people use it, the economics of behavioral advertising break down, creating the space for real regulation and protection of vulnerable populations.