HiddenLayer: Chris Sestito on ML/AI Security Incidents and Defense Capabilities
Apr 4, 2023
auto_awesome
Explore real-world attacks on machine learning models, AI security capabilities, and market demand for security tooling with Chris Sestito. Learn about HiddenLayer's journey, monitoring machine learning for security, and tools for defending against threats in ML/AI systems.
Adversarial machine learning attacks can significantly impact threat detection models, leading to the need for dedicated security solutions for AI.
HiddenLayer's Model Scanner helps organizations scan for malicious AI models in repositories and mitigate risks of deploying compromised models.
Deep dives
Evolution of Hidden Layer and Birth of Idea
Chris Sestito, CEO of Hidden Layer, delves into the evolution of the company starting from the vital incident at Silence where an adversarial machine learning attack significantly impacted their threat detection model. This incident sparked the realization for the necessity of a dedicated security solution for artificial intelligence. The journey of Hidden Layer commenced with the recognition that the market was becoming more conducive for such a solution, leading to the establishment of the company.
Challenges in Traditional Security Stack
Sestito highlights the vulnerability of organizations unintentionally downloading compromised machine learning models from open repositories due to supply chain issues. The lack of security built into AI model repositories poses a significant risk. Hidden Layer's Model Scanner addresses this threat by allowing organizations to scan for malicious models and mitigate risks of deploying compromised models.
Monitoring and Response Capabilities
The flagship product of Hidden Layer, MLDR, offers both detection and response functionalities. It empowers organizations to respond to adversarial machine learning attacks in real time with options like rate limiting requesters, cutting off access to models, or rerouting to a model honeypot for intelligence gathering. It also facilitates tracking and remediation of compromised models.
Future Outlook and Growth
As the AI security landscape evolves rapidly, Hidden Layer's proactive approach enables them to stay ahead in the market. The company is actively seeking investments and expansion opportunities to further strengthen their cutting-edge solutions. With a focus on industry growth and staying at the forefront of AI security, Hidden Layer remains a pivotal player in this dynamic space.
CEO and Founder of HiddenLayer, securing organizations building or using machine learning models
Previously VP of Engineering at Qualys
Previously Senior Director of Data Science at Agari
Previously Director of Threat Research at Cylance when they were hit by a model inference attack back in 2019
Fun fact: The HiddenLayer co-founders have worked together for the last five years!
Check out the episode for our conversation about real-world attacks against machine learning models, the current state of AI security capabilities including monitoring and scanning, and the market appetite for this tooling.