FIU researchers develop blockchain defense against AI data poisoning attacks

Florida International University researchers have developed a blockchain-based security framework to protect AI systems from data poisoning attacks, where malicious actors insert corrupted information to manipulate AI decision-making. The technology, called blockchain-based federated learning (BCFL), uses decentralized verification similar to cryptocurrency networks to prevent potentially catastrophic failures in autonomous vehicles and other AI-powered systems.

What you should know: Data poisoning represents one of the most serious threats to AI systems, capable of causing deadly consequences in critical applications.

• Dr. Hadi Amini, an associate professor of computer science at FIU, demonstrated how a simple green laser pointer can trick an AI camera into misreading a red traffic light as green with over 80% probability.
• “Data poisoning is a type of cyberattack that can happen at different stages of developing an AI system,” Amini explained, noting it can affect everything from drones to self-driving cars.
• The attack exploits AI’s fundamental dependence on massive datasets for learning and decision-making.

How the solution works: The BCFL framework creates multiple layers of verification by leveraging nearby devices to cross-check AI decisions.

• “Instead of solely relying on the camera of your car, we are using the cars or autonomous systems around you,” Amini said. “All of these cameras should agree on the decision, whether it’s a green light or red light.”
• Ph.D. candidate Ervin Moore compared the system to sports referees: “If one referee finds a foul, all the referees congregate and look over the information. That’s essentially what the system does to verify the data.”
• The blockchain technology adds decentralized verification layers that can identify and reject poisoned data before it reaches decision-making processes.

The research team: A 10-person team of FIU Ph.D. students is working under Amini’s leadership to develop and test the security framework.

• The team is currently partnering with industry collaborators to test BCFL in real-world scenarios.
• They expect the technology to be ready for full deployment within the next couple of years.

Why this matters: While autonomous vehicle companies already have security protocols in place, the FIU team’s approach offers an additional defense layer against increasingly sophisticated cyber threats targeting AI systems that could have life-or-death implications.