Why BlindAI?
AI models are everywhere. Recent releases, like GPT4, have skyrocketed adoption and democratized their use in all industries. We believe this is great for progress, but that we need to take action to develop solutions that protect confidential data when using those models. Take ChatGPT's example: thousands of employees used it - unrestricted - to help them in their tasks, feeding it company data that could then be viewed by any employee working on improving the next version of the model.
This is why we made BlindAI, so AI models can be used while preserving users' data privacy (and without putting the model at risk). To understand the advantages you'd gain in using us, here's a little tour of the current options used to deploy AI models.
Don't choose between ease-of-use and privacy
AI solutions to be deployed in one of three ways: on the Cloud, on-premise or on-device.
Cloud deployment offers users ease-of-use and a wide offer of AI models but it puts users' data at risk. Their privacy is left in the hands of the person or company who operates the service, the cloud operator, their sysadmins and other people who each need to be trusted. In general, this means that Cloud deployment is currently not suitable for users with sensitive data such as hospitals wanting to leverage patient data to improve healthcare.
Users with sensitive data may instead turn to on-premise or on-device deployment as a more privacy-friendly alternative to Cloud deployment, but this increased privacy comes at a great cost in terms of ease-of-use. - On-premise deployment requires on-site expertise that is not available to many companies. - AI models storage space and power are limited by on-site hardware (on-premise deployment) or user devices (on-device deployment). - AI models need to be embedded in the application distributed to the client, putting them at risk of being stolen in on-device solutions.
BlindAI offers the best of both worlds: the ease-of-use and wide offer of Cloud deployment with the increased privacy of on-premise and on-device solutions. We take advantage of the power of confidential computing, and more specifically Intel Software Guard Extension (Intel SGX), to enable user data to be processed remotely without the risk of unauthorized access or tampering.
Example: AI integration for hospitals
Imagine you have a hospital wanting to use AI to help doctors be more efficient. Doctors often take voice notes in a rush, then need to transcribe them into writing for later reporting. This takes a lot of time and it could be automated with AI speech-to-text.
The hospital doesn’t have the expertise, the infrastructure or the time to do so. If they want to implement it, they would turn to external help - for example an AI provided through the SaaS of a startup. This would be beneficial because there would be no cost for onboarding or for maintenance.
The issue is that, when sending data to that startup, no technical guarantee can be provided that the confidential medical information contained in those audios will not be compromised. Data can be encrypted in transit, and even if the startup uses a secure Cloud, it can still be put at risk because it will be in clear in memory before being analyzed for transcription by the AI model. Therefore the hospital has to entrust their data to this startup which might not have put in place the proper security measures to ensure data confidentiality.
This key security flaw often cuts off this type of collaboration from the get go. The hospital would likely refuse to send their medical data to the AI startup. Being able to use speech-to-text could have saved many hours for the hospital’s personnel and benefit the startup, but healthcare data security is just too sensitive to take any risk.
This is where BlindAI comes in. Using BlindAI Core, the AI company could serve their AI model in a highly isolated computing environment, a Trusted Execution Environment, which ensures no-one, including their employees, can access the hospital's data. This allows this kind of collaboration to go ahead to the great benefit of both parties.
Next steps
To get see a hands-on demo of how you can use the BlindAI API, check out our Quick Tour!
To find out more about the Confidential technologies behind BlindAI, check out our guide to Confidential Computing.