Assaf Pilo - AI as an Enabler for Detecting PI Across Large Data Infrastructures

So I think, yeah, two presentations and beers and wine probably is waiting for you in the other room. So I'll try to keep it short. I'm sure that some of the challenges within some of the challenges that you all see you, you could find these three ones migration to the cloud cloud, adoption, data, privacy, and overall looking at data and especially dark data, understanding what kind of documents you have, what kind of sensitive information or data do you have?

Do you have PII, any business critical, any intellectual property that's hiding within your maybe millions or tens of millions or hundreds of millions of, of documents. And it can be across platforms, whether it's cloud on-prem SMB, SharePoints, different providers, and, and it's across all different business units, or most of your business is done using documents. If we're looking at anything from contracts, agreements, price, quotes, POS SLAs, even your product definitions and roadmap is, is with documents. So we're looking at business critical data.

We're looking at data that is sometimes obsolete that we can archive or delete. We're looking at all the data that we're not really sure what's what's there. And there are different with regards to PI. There are many kinds of personal information pieces or, or personal information that we want to detect that we use within different contexts.

So if it's health information financial, if it's social, if it's any type of identifier, the main thing is really understanding what is the context of that PI and the context influences not only regulation and you know it from GDPR, ACP and other ones, but also what we, what you wanna do with the data. How do you want to handle the data? Do you need to keep it, do you need to store it in a certain way? Do you have a retention policy with regards to that data, et cetera, et cetera. So some of this data, you can apply minimization methods so that you can delete obsolete data.

You can find duplicates, you can find documents that you're not using anymore. You need to segregate data based on geolocation or business units. HR data should be with HR and legal should be with legal. If I have employees that are reporting to me, I probably have their CVS and contracts somewhere. That's not good. And I need to find all these rogue documents and I need to move them to where they belong.

When we, when we collaborate With Documents, we need to be able to understand, Do we Have our labeling? Correct? Do we have our protection the Right way? Do we have the privileges of Accessing different Follow repositories set up In the correct way? Obviously You all know D SARS and right to be forgotten. How do we identify people in hundreds of millions of documents, Both accurately And in a way that we can then create actionable items from these results, As well as What do we do when we have conflict?

If I have a contract that I encrypted and I have Great Security measures with This Contract, maybe I have some copies that I haven't Protected. Can I, can I find those copies? Do I want to know about those Copies That include this probably the same information as the, as the encrypted one?

So All these challenges bring us down to using AI machine learning automation, because when we look at A repository with 500 million documents, it's almost impossible to, To Even approach it and start classifying it with human labor Or With any, any other form, rather than having automation and having multidimensions running on every each and every analysis or scan. So that will, you'll be Able to point at Those specific Documents, that Specific data That you need to Take Actions with.

Again, it can be move Archive, protect, delete label Encrypted, and, and the different dimensions Are across that big data across all those big repositories And take into consideration many parameters. So it's Taking some Cluttered Room and making, make, putting it in boxes mat, putting it in clusters and in classes that we can, we can move, or we can work with Just Being aware of the time.

Oh, two minutes, four minutes. Okay, great. So our approach is, Is, and, and the approach that we are using is machine vision.

It's creating an object out of each and every document that we can then compare To every other Document in the organization, analyzing it deeply into the bite Code level, Analyzing the text and the context across multiple Repositories, so That we can Create both out of the box Recommendations for clustering that data, As well as refer To your specific business needs in order to comply with regulations in order to address your sensitive data in your organization.

And it, it really all comes down to cost because if we would have endless amount of people being able to read files and compare them and understand what's what's in them, we wouldn't need tools, but when we are referring to big data and the growth of documents in every organization is exponential. We see documents, we see customers, mid-size customers with tens of millions and then growing to the hundreds of millions of documents. And it's extremely difficult to have that spent using human, the human factor and using labor in order to do that.

So AI comes in, in all the different stages, having a centralized repository for clustering files and unstructured data, which is, is all around the different documents. The, the business that you do that is representing those, represented those in those documents, categories, being able to, to put the data in categories that we can then analyze what we wanna do with them. Do we want to minimize the data?

Do we need to segregate the data, analyzing where the sensitive files are, what we need in order for our business to run what we need so that compliance and, and the regulation will be happy and triggering whatever actionable items or whatever actions we need to take in order to put our data in order for moving to the cloud for complying with GDPR or other privacy regulations, for being able to understand where the risk is and reduce our attack surface by securing our data better, knowing what the data actually, what, what sensitive data and what the da, where the data actually is.

Thank you so much. I think I'm right on time.