Ethereum co-founder Vitalik Buterin and Ethereum Foundation AI lead Davide Crapis shared a blog post explaining that users can interact with large language models privately and prevent spam and cheating via zero-knowledge proofs.
Ethereum developers build private way to pay for AI chatbots
Vitalik Buterin and Davide Crapis say AI chatbots raise serious privacy concerns today because users share personal and sensitive information via API calls that can record, track, and sometimes connect those requests back to the owner.
The developers of these chatbots say they can’t ignore the issue any longer, because the risk of personal data exposure keeps growing as people use AI every day.
Because of this, Buterin and Crapis explain that AI providers can either ask users to sign in with an email address or pay with a credit card, or use blockchain payments for anonymity.
If companies settle on email addresses and credit card payments because they’re more familiar, users’ privacy will be at risk, as every chatbot request links to someone’s real identity. This can lead to profiling, tracking, and even legal risks if people present these logs in court.
For blockchain payments, users would have to pay on-chain for every request, but the process is slow and costly, and it creates a visible record of every message. Privacy when paying per request will be impossible again because the user’s transaction history will be easy to track.
Ethereum developers are now proposing a new model in which a user deposits funds into a smart contract once and then makes thousands of private API calls. This way, the provider is sure the requests have been paid for, and the user doesn’t have to confirm their identity every time they interact with the chatbot.
Buterin and Crapis say the new model will go a long way toward keeping people safe while allowing the technology to grow.
Zero-knowledge proofs stop bad behavior without revealing user identity
Ethereum developers say the system will use zero-knowledge cryptography to prevent cheating and abuse because it allows a user to prove something is true without revealing their identity. Vitalik Buterin and Davide Crapis explain that zero-knowledge tools will help honest users remain anonymous while exposing bad actors who try to break the rules.
The new model will use a tool called Rate-Limit Nullifiers (RLN), which allows users to make anonymous requests and catch anyone who tries to cheat the protocol.
This process begins when an account owner generates a secret key and adds funds to a smart contract, which is then used as a buffer for API calls. The account owner will fund the account once and then make private calls using the funds deposited, rather than making separate transactions each time they make an API call.
This is an obvious limitation because an individual can make only as many calls as they deposited money for. Then, every time the user makes a request, the protocol assigns it a ticket index, and the user must produce a special proof, called a ZK-STARK, that they are still spending funds deposited with the protocol, as well as any refunds they are entitled to. At the same time, the system also processes refunds, as AI requests are not always of equal cost.
The protocol also generates a unique nullifier for each ticket to prove usage and immediately identifies attempts to reuse the same ticket index for two different requests.
According to Buterin and Crapis, abuse is not only double-spending, since some users may try to break the provider’s rules by sending harmful prompts, jailbreaks, or requests for illegal content such as weapon instructions.
The protocol thus adds another layer called dual staking, where the user’s deposit is subject to strict math rules, and the other is subject to provider policy enforcement.
