The Israeli Delegation to ITC 2022 | DeepKeep: We develop an open and flexible platform to easily and seamlessly digest multiple new attacks and protection algorithms


From left to right: Yossi Altevet CTO, CEO Dr. Rony Ohayon and Guy Sheena CBO

DeepKeep is developing an innovative automated platform for securing AI models and datasets from various adversarial AI attacks such as evasion, stealing and poisoning and mistakes. The platform offers a comprehensive set of AI security tools including penetration testing, online/offline attack detection and a diversified protection suite. The set of tools provided by DeepKeep’s platform can be categorized into the following four pillars: Risk Analysis, Robustification, Attacks Detection and Attack Mitigation. Artificial Intelligence (AI) models are inherently vulnerable to adversarial attacks. According to recent market studies, 72% of organizations have experienced security breaches or threats in their AI systems.

The key figures are – CEO, Dr. Rony Ohayon has over 20 years of entrepreneurial, developmental, technological, academic, business, and managerial experience in the high-tech field. The CBO, Mr. Guy Sheena is experienced in taking a new product into new markets including market education, partnering, and selling. The CTO, Mr. Yossi Altevet has rich experience in various domains including IoT, Security, Automotive, telecommunication, etc.

This article is a part of a series of articles presenting the Israeli delegation to the ITC conference.

What is your business model?
The business model is built in layers: 1st layer is the domain (video, Image, Voice, tabular); 2nd layer is a type of solution (service/offline/online); & the last layer is the suite type (Basic/advanced/premium). The company plans to charge its customers an annual fee that will include risk assessment, firewall, prevention and detector per model.

How do you see the current situation in that area on the insurance industry today?
The BSFI and IoT industry is shifting towards decision support systems and automation. Many of the tasks that have traditionally been performed by humans prove to be relatively easy to automate by training of suitable AI models to make the decisions. Cutting human assessment out of the loop can bring substantial savings in manpower, expedite assessments, provide improved customer experience and reduce insurance rates. Despite their relative accuracy and efficiency, the same accurate and efficient AI models that can assess car damage, property value, potential customer credibility, etc., are susceptible to adversarial attacks that will no doubt bring upon fraud, financial damages, and confusion. It is most likely that malicious actors perform such attacks today and insurance companies are not even aware of the extent of the financial damages that they suffer due to these attacks. Protecting AI models for insurance companies is an inevitable mandatory step in the development, deployment, and real-time operation of their services.

What investments have been made in the company so far?
The company raised Pre seed 1,450k USD from private investors & Israel Innovation Authority in 2021. In 2022, DeepKeep has completed Seed round of $4M from private VC.

 What is your current business situation?
DeepKeep is presently developing its technology and at the same time seeking design partners in major players in the three verticals that we are targeting (Fintech, Automotive, and Security) and are engaging with Proof of concept (POC) implementation projects that feed back into our development and adjust our offering to the market stage and ability to digest our solutions.

What is your competitive advantages?
We develop an open and flexible platform to easily and seamlessly digest multiple new attacks and protection algorithms which are introduced by academic research very rapidly.

We develop our platform to provide non-intrusive value by sniffing and parallel processing, which provides detection and alerts while not interfering and/or compromising the security of standard production processing.
We design and build our platform for the different actors throughout the AI lifecycle.
We design and build our platform to operate on Blackbox models access remotely via URL as well as on leveraging customers’ computing power.
We collaborate with the academy in multiple efforts to develop innovative SOTA attacks.
We have built a winning experienced executive and technical team that will know how to navigate this effort and steer the company to success in these upcoming stormy waters.



Leave A Reply