Imagine a hospital wanting to use artificial intelligence to analyze patient records but needing to ensure the data stays private.
A special encryption method, called homomorphic encryption, allows calculations on encrypted data without ever decrypting it.
This means that even the company processing the data never sees the original information.
While this sounds like the perfect security solution, homomorphic encryption has a major problem—it is extremely slow and difficult to use in real-world applications.
But now, researchers at MIT have found a new way to make this process simpler and more efficient.
A simpler, faster approach
MIT researchers have developed a new theoretical encryption method that makes secure computing on encrypted data much easier.
Instead of relying on complex and slow encryption techniques, their approach combines two simple cryptographic tools to create a more powerful system.
This new method is known as somewhat homomorphic encryption. While it doesn’t allow unlimited calculations on encrypted data like fully homomorphic encryption, it allows a useful number of operations while keeping the data secure. This could be enough for applications like private database searches, secure AI processing, and encrypted statistical analysis.
How does it work?
Encryption works by adding noise to data, making it unreadable. However, every time a calculation is performed, the noise grows. If too many calculations are made, the noise becomes so overwhelming that the original data gets lost.
The MIT team found a way to control this noise by limiting operations to a specific class of functions. Their method allows many additions but only a few multiplications, preventing too much extra noise from being generated.
To do this, the researchers combined a simple encryption method that supports addition with a new cryptographic assumption. This combination unexpectedly resulted in a stronger and more flexible encryption method than either approach alone.
The encrypted data is stored in a matrix format, and performing operations is as easy as adding or multiplying the matrices. Mathematical proofs confirm that this method keeps the data secure while allowing useful computations.
Although this encryption method is still theoretical, it brings researchers one step closer to practical homomorphic encryption. The next challenge is to make it fast enough for real-world use, such as secure AI processing, cloud computing, and privacy-preserving applications.
Henry Corrigan-Gibbs, one of the lead researchers, explains, “The dream is that you could ask an AI like ChatGPT a question, encrypt it, and receive an answer—all without the AI ever seeing your original request.”
While fully homomorphic encryption is still far from practical, this new method offers hope for a future where privacy and AI can work together. As researchers continue refining the technology, it could lead to a major breakthrough in data security, protecting sensitive information in new and powerful ways.
Source: MIT.
