India has proposed letting AI makers train on hospital and bank data without ever handing that data over. Here is how the plan works and why businesses may care.
India's government has floated a plan to let companies build AI from sensitive local data, such as hospital and bank records, without the data being handed over. Its chief science adviser put it this way: "the model travels to the data, rather than the data travelling to the model." The plan is called DEPA for AI Training.
DEPA stands for Data Empowerment and Protection Architecture. It comes from India's Ministry of Electronics and Information Technology, known as MeitY. Organizations that hold data, such as hospitals, banks and government agencies, would share anonymized and encrypted data with a secure exchange platform. AI builders would then train their models there. The data owners set the rules for how their data can be used, and the platforms enforce them.
The framework describes secure computing environments built on hardware, where data stays encrypted and isolated even while it is being used. It also calls for techniques meant to stop a model from learning or revealing details about any one person. Every action inside these environments is logged in a way that cannot be changed, so auditors can check it later.
One motive is control. India thinks offshore model makers will not be motivated to address the country's needs or keep data owners in charge. The government also lists hoped-for benefits:
These are the government's hopes, not results. The framework also ties DEPA to India's broader push for publicly funded digital tools, such as its payments system, meant to keep Big Tech from dominating the local market.
India already runs several data exchanges. It is unclear how much work remains before DEPA is a working reality. The material gives no launch date, and it does not say which organizations have agreed to take part.
Any company that holds customer records may eventually be asked how it could use that data for AI without giving it away. India's approach is one early answer: keep the data in a locked room, let the model visit, and keep a record of everything that happens.
India's plan sends AI models to locked-down data instead of copying data out, an idea worth watching if your business holds customer records.
A former Nvidia product manager has started Udu Technologies to let African startups and institutions rent AI computing power instead of buying it.
Read more: Ex-Nvidia engineer starts company to bring AI computing power to AfricaGenRobot AI raised $100 million to record how people work, so robot makers can use those recordings to teach machines home, factory and warehouse tasks.
Read more: Beijing startup raises $100 million to teach robots by watching peopleRecruiters say AI now handles much of the screening work, so employers may start asking what a fee of a share of a new hire's salary actually pays for.
Read more: AI is speeding up hiring and putting recruiter fees under pressure
