We train our AI on a myriad of different smart contract models written in Solidity and Haskell and have developed a search space large enough to understand most fatal security vulnerabilities to automatically patch them.

Example

In this function a malicious user will be able to extract several times his own Ether balance from the contract before updating the internal balance of that user. Which means, stealing all the funds of the bank in one function call.

Process

AI identifies vulnerability, will parse through a search space to find a test suite of solutions to show the developer. The developer will pick one and the AI will automatically modify the source with the patch.

Solution

Set the balance of the user to 0 and then send the corresponding Ether balance using the more secure function `msg.sender.transfer()`. That way the balance of the user will be zero when you try to extract it again from the same function call.

Haskell and Solidity

We intend on initially focusing on Haskell and Solidity written Smart Contracts. We will eventually start training with other languages to start catering to all security vulnerabilities.