BY DUNCAN RILEY
Artificial intelligence and machine learning systems cybersecurity startup Protect AI has today emerged from stealth mode with $13.5 million in new funding and its first product, NB Defense.
NB Defense is a free product that’s claimed to be the industry’s first security solution to address vulnerabilities in a core component used at the beginning of the machine learning supply chain: Jupyter Notebooks. They’re web-based applications that allow developers to create and share documents containing live code, equations, visualizations and other data for coding purposes, including data cleaning and transformation, statistical modeling, data visualization and machine learning.
As of today, there are believed to be more than 10 million publicly accessible Juypter Notebooks, with the figure growing by more than 2 million annually. There are also believed to be many more Juypter Notebooks installations in private repositories.
Protect AI was founded by a leadership team with AI business experience working for Amazon Web Services Inc. and Oracle Corp., including co-founder and Chief Executive Officer Ian Swanson, who was previously worldwide leader for AI and machine learning at AWS. The round was co-led by Acrew Capital and boldstart ventures, with Knollwood Capital, Pelion Ventures and Avisio Ventures also participating.
“I have seen more than 100,000 customers deploy AI/ML systems and realized they introduce a new and unique security threat surface that today’s cybersecurity solutions in the market do not address,” Swanson said in a statement. “This is why we founded Protect AI. ML developers and security teams need new tools, processes, and methods that secure their AI systems.”
Swanson explained that given that nearly all machine learning code begins with a notebook, the company thought that it was the most logical place to start to accelerate a needed industry transition.
“We are launching a free product that helps usher in this new category of MLSecOps to build a safer AI-powered world, starting now,” Swanson added. “But we have many more innovations that will be released quickly across the entire ML supply chain.”
As MLOps has helped increase the velocity of machine learning being used in production, opportunities for security incidents have increased and new vulnerabilities have been created in the enterprise machine learning supply chain. Some of the security risks include Jupyter Notebooks incompatible with existing static code analyzers, arbitrary code execution in serialized models, poisoned training data and model evasion using adversarial machine learning techniques.
NB Defense creates a translation layer from traditional security capabilities to enable scans of Jupyter Notebooks, then communicates findings natively in the notebook or via reports with context-specific links to problematic areas within the notebook for remediation.
The offering scans a notebook to check for the standard Common Vulnerabilities and Exposures database in open-source machine learning frameworks, libraries and packages, application tokens and other credentials, as well as nonpermissive licenses in the frameworks, libraries and packages.
NB Defense is available today under a free license. Users can install NB Defense and use the JupyterLab Extension or Command Line Interface.