Prague based Resistant AI has nabbed a $2.75M seed round. The security startup’s machine learning technology is designed to be deployed on top of AI systems used for financial decision making to protect customers in markets such as financial services and ecommerce from attacks such as targeted manipulation, adversarial machine learning and advanced fraud.
The seed round was co-led by Index Ventures (Jan Hammer) and Credo Ventures (Ondrej Bartos and Vladislav Jez). Seedcamp also participating, along with Daniel Dines, CEO of UiPath; Michal Pechoucek, CTO of Avast and other unnamed angel investors. Bartos joins the board of directors on behalf of the investors.
The startup sells an additional layer of protection that’s specifically designed for tightening security around automated functions such as credit risk scoring and anti-money laundering by using tech to detect fake documents that feed such systems. Its tech is also aimed at uncovering suspicious patterns of transactions which might indicate a strategic attack on the model itself or an attempt to copy sensitive data.
“Historically, all systems that make high-value financial decisions become targeted. This is already happening with the automated systems deployed by our fintech and financial customers and we are here to protect them,” said Martin Rehak, founder and CEO, in a statement.
The seed round is Resistant AI’s first tranche of external funding, with the founders bootstrapping the company since starting up in February 2019.
“We have onboarded the first customers in 2019 and the funding will help us scale our sales organisation to meet the rising demand from banks and fintechs,” Rehak told us. “We are protecting the AI&ML systems used in financial automation from manipulation or misuse by smart attackers.”
Resistant AI has two products it offers its customers at this stage: First, document inspection. It offers a machine learning system that’s designed to flag and reject “malicious documents” submitted for automated processing. “Bank statements, payslips, invoices, purchase orders and KYC documents submitted to fintechs and banks are frequently manipulated or completely falsified,” explained Rehak. “Resistant Documents, our first service, identifies and rejects the suspicious or malicious inputs.”
A second offering — Resistant Transactions — applies AI to spot problematic transaction patterns.
“We work with the fact that most attacks on AI systems require extensive interaction to discover the vulnerability,” he said. “Our system is unique by inspecting all the customer queries (which can take form or payments, money transfers or credit applications assessed by the system we protect) in context of similar queries. By looking at the stream of queries statistically, we can recognise and block the attacks that seek to steal the information embodied in the model (information stealing) or, worse, aim to nudge the system into making the wrong decision by exploiting an existing bias in the system.”
Resistant AI isn’t breaking out customer numbers yet but Rehak said it onboarded its first customers last year. “The funding will help us scale our sales organisation to meet the rising demand from banks and fintechs,” he added, saying also that it will be spending on building out product features and extending functionality, as well as on beefing up the sales and go-to-market team.
“Right now, our target customers are financial and fintech startups, as well as other companies deploying the automated process (both software and RPA) in their financial processes,” he added. “The financial systems are our current focus, but the attacks on machine learning are relevant in many other areas: process automation, e-commerce, manipulation of ‘trend detection’ algorithms in social media and other opportunities.”
It’s using a SaaS model — preferring a value approach to pricing, per Rehak. “Our problem and approach is new, and we feel that the value pricing model aligns the incentives between us and the customer in the optimal way,” he said on that.
Asked who he sees as the main competitors for the business, he cited Google Brain plus the tech giant’s activities in adversarial machine learning.
The majority of work in this area is currently done in-house by the large tech companies building their own proprietary systems — such as Google and Microsoft, he added.
Other competitors he mentioned were Inpher, which is enabling machine learning on encrypted data; Sentilink, which is doing detection of synthetic identities in the US; and Bullwall (Denmark) and YC-backed Inscribe (US/Ireland) which are focused on document forgery.
Resistant AI’s founders have a background in machine learning applied to cyber security problems having founded Cognitive Security, an earlier startup which they subsequently sold to Cisco in 2013. Over some 12 years working in the security industry Rehak said they saw how attackers targeting AI systems were getting increasingly sophisticated in avoiding detection — which gave them the idea for their latest business.
Commenting on the seed funding in a statement, Jan Hammer, general partner at Index Ventures, added: “Automation, efficiency and reliability are cornerstones of financial innovation. As machine learning takes more and more nuanced financial decisions, it needs to be protected. And this is not true only in finance, but the attacks will rapidly spread to other domains as well. More of our activity today takes place online, a trend accelerated by COVID-19, and one we believe will last. With criminals ready to take advantage of every vulnerability, the need for solutions such as those from Resistant AI has never been greater.”
via
Tingle Tech
