Guest Post: A Cryptography Boom Following the Era of AI

Will homomorphic encryption enable an AI-secure future?

I’m Maxim, a rising junior at Stanford studying CS, and this summer I’m working at Also Capital as a Scout Fund investor.

A fellow student and I recently discussed what we thought would be the 'highest-potential markets' in the wake of the AI boom. We talked about it in the context of a number theory cryptography class I plan on taking next quarter called Math 110.

My bet is on homomorphic encryption.

As Savvas Savvides explains, homomorphic encryption (HE) allows computations on encrypted data without needing to decrypt it first by transforming the original data into an encrypted equivalent while preserving the relationships between the elements. This ensures that the results are consistent with operations on the original information. It's like hiring a private chef, blindfolding them, giving them a bunch of ingredients, and telling them to cook you something—they’ll still make a meal, but they won’t necessarily know what you put in. The goal is to have AI models operate on encrypted data where only the key-holders can decrypt the output without exposing sensitive information. 

For example, if the government wanted to use OpenAI’s ChatGPT services on a classified case—obviously, they can’t. Or, asking Anthropic’s Claude whether your current checking account password is strong enough is a big security risk. These are problems we can solve! 

Given the pattern we are seeing of diminishing personal privacy, coupled with the data-hungry nature of ML algorithms, the future of our sensitive information is at greater risk than ever before. Tensions building overseas concerning the race for AI with competitors like China only add fuel to the fire. With homomorphic encryption, we can use LLMs without giving them access to our sensitive information.

Simply put, the need for even better encryption methods will naturally skyrocket, meaning that an AI revolution is also a delayed cryptographic revolution. HE might not be the key to success, but any progress in related fields is a step toward a future led by AI with zero compromise on our privacy and security—unlimited power with no drawbacks, the way Emperor Palpatine wanted it.

If you have any experience or knowledge of cryptography, and/or know of awesome startups that are working on difficult data encryption problems, please reach out to me via maxim@alsocapital.com or LinkedIn! I’d love to connect!