Breaking Through the Italian ChatGPT Ban with OpenAccessGPT

Back to Friday 31st of February, Italy's Data Protection Authority (the so called Guarantor) temporarily banned OpenAI's ChatGPT chatbot and launched a probe over a suspected breach of the artificial intelligence application's data collection rules.

The Guarantor have accused OpenAI of neglecting to check the age of ChatGPT users and engaging in massive collection and storage of personal data without any legal justification for training the chatbot.

In response, OpenAI has halted Italian user access to the chatbot and plans to offer solutions to address these concerns.

Garante clarified that it is not attempting to hinder AI progress but rather emphasizes the necessity of following regulations to safeguard the privacy of Italian and European citizens.

In the meanwhile, in response to the ChatGPT block in Italy, we have seen many people resort to unreliable VPNs or non-transparent apps to continue accessing the service.

So, during the Easter holidays, I worked on a side-project with my colleague Diego Pizzocaro to offer an open-source alternative to ChatGPT that allows anyone to be in control of its own data without needing a VPN in Italy.

It started from an existing Github project available under MIT license called chat-with-gtp. It got our attention because it was an open-source project, consisting of a React frontend app and a Node.js backend app, available both on-line and via a Docker instance.

We forked the project and adapted it to the Italian context, providing Italian localisation and getting rid of all the backend (and Dockerization) part, to provide just the React interface, that has the characteristic of relying only on informations stored within the browser context.

We called it OpenAccessGPT and we released it both freely on Vercel for an immediate usage and also as an open-source project that can be installed in any workstation.

OpenAccessGPT has the important characteristic of using OpenAI APIs and stores conversation data exclusively locally. No VPN required from Italy, as communication via API is not subject to investigation by the Privacy Guarantor.

The fact that OpenAccessGPT uses the OpenAI APIs puts on the table two important aspects that come closed to some of the things that the Italian Guarantor is looking for.

In specific, as per it is stated in the API data usage policies of OpenAI:

1. OpenAI not uses data submitted by customers via API to train or improve its models, unless the user explicitly decides to share its data with OpeAI for this purpose.

   
   

2. Any data sent through the API is retained for abuse and misuse monitoring purposes for a maximum of 30 days, after which it will be deleted (unless otherwise required by law).

Hence, to summarise, OpenAccessGPT takes advantage of the new OpenAI APIs for ChatGPT, which only store data for 30 days on OpenAI servers and then are permanently deleted, without using them to train and improve OpenAI AI models.

Once we released OpenAccessGPT we started to spread the voice and in only one day we reached more that 300.000 people that expressed their support and started to play with the tool and asked how to collaborate to the project.

We were emotionally touched by all of this, confirming how often from tiny but clear ideas a meaningful impact can be bootstrapped.

Many open-source LLM projects are emerging, which can also be installed directly on your PC, and we believe this is a trend destined to grow, in line with our OpenAccessGPT project - and we hope one day to include many other AIs to interact with.

We are sure that many other developers, hackers and technology enthusiasts in Italy are working "silently" to find alternative solutions to access these AIs - with due attention - and generally advance this incredible technology and its adoption, which it's changing the world. If any of them would like to join us and give us a hand we will be very happy!

If you are interested in helping us or trying it, do not hesitate to join the project's Discord Server and participate in the GitHub project. Any other useful info is also present on our Project website!