HiddenLayer Secures $50 Million to Enhance AI Security Solutions
HiddenLayer, an innovative startup dedicated to safeguarding AI systems from adversarial threats, announced today that it has successfully secured $50 million in a funding round co-led by M12 and Moore Strategic Ventures. Additional contributions came from notable participants including Booz Allen Hamilton, IBM, Capital One, and TenEleven. This latest round brings HiddenLayer’s total funding to $56 million.
The newly acquired funds will be used to bolster HiddenLayer’s go-to-market strategy, expand its workforce from 50 to 90 employees by the end of the year, and enhance its research and development efforts, as co-founder and CEO Chris Sestito shared in an email.
“HiddenLayer is a cybersecurity company focused on defending AI from adversarial attacks. Specifically, we extend detection and response capabilities to AI systems,” Sestito stated. “We are rapidly scaling to meet the increasing global demand for our machine learning security platform across diverse industries.”
Founded in 2019 by Chris Sestito, Jim Ballard, and Tanner Burns, HiddenLayer was born out of Sestito's previous role leading threat research at Cylance, an antivirus firm acquired by BlackBerry.
The platform provides essential tools to shield AI models from adversarial attacks, vulnerabilities, and malicious code injections. It actively monitors AI system inputs and outputs, ensuring model integrity is tested before deployment.
“Many data scientists depend on pre-trained, open-source, or proprietary machine learning models to accelerate analysis and streamline testing,” explained Sestito. “This reliance on publicly available models can expose organizations to transfer learning attacks, where malicious actors tamper with these models.”
To alleviate concerns about potential access to proprietary models, HiddenLayer employs techniques that focus exclusively on vectors—mathematical representations of input and output data—without compromising client confidentiality.
“Our system autonomously learns what is considered normal for each unique AI application without requiring manual instruction,” Sestito added.
Furthermore, HiddenLayer contributes to the MITRE ATLAS, a comprehensive knowledge base detailing adversarial tactics and techniques. Sestito emphasizes that HiddenLayer can counter all 64 unique attack types outlined in the ATLAS, which includes threats like intellectual property theft, model extraction, inferencing attacks, model evasion, and data poisoning.
Despite uncertainty over the groundbreaking nature of HiddenLayer’s solutions, industry expert Mike Cook from the Knives and Paintbrushes collective acknowledges the value in making knowledge about AI attacks more accessible through their platform.
Quantifying actual instances of large-scale attacks on AI remains challenging. Research interest in AI security has surged, with over 1,500 papers published in 2019 on Arxiv.org, up from just 56 in 2016, according to Adversa. However, public records detailing hacking attempts on systems such as commercial facial recognition software are scarce.
Nevertheless, some government agencies are raising alarms about potential threats to AI systems. Recently, the U.K.'s National Cyber Security Center cautioned against the manipulation of technologies behind large language model chatbots like ChatGPT, citing risks of unauthorized data access, generation of harmful content, and unintended consequences.
The U.S. Government's Office of Science and Technology Policy also released an "AI Bills of Rights," advocating for pre-deployment testing, risk assessment, and ongoing monitoring of AI systems to ensure their safety and efficacy.
Businesses are increasingly adopting this perspective as well. In a Forrester study commissioned by HiddenLayer (though this warrants some caution), a significant majority of responding companies revealed that they currently rely on manual processes to tackle AI model threats, with 86% expressing serious concerns regarding their machine learning security.
Gartner's 2022 report indicates that nearly two in five organizations experienced an AI privacy breach or security incident in the last year, with 25% of these incidents being malicious attacks.
Sestito asserts that regardless of the current landscape, the threat to AI systems is poised to expand alongside the growing AI market—benefiting HiddenLayer in the process. He recognizes that several startups, including Robust Intelligence, CalypsoAI, and Troj.ai, already provide solutions to enhance AI system resilience, yet claims that HiddenLayer's AI-driven detection and response strategy is unique.
HiddenLayer has gained notable traction, boasting partnerships with Databricks and Intel, and serving Fortune 100 clients across various sectors, including financial services, government, defense, and cybersecurity, with clients like the U.S. Air Force and Space Force.
“The rapid pace of AI adoption has left organizations struggling to implement adequate processes, personnel, and controls to safeguard against the inherent risks and threats associated with machine learning,” Sestito remarked. “As the risks of integrating AI continue to rise, we are scaling quickly to meet the demand for our platform across diverse industries globally.”
Ranking
