Check Point Software has announced plans to acquire Deepchecks, a startup that boasts a team led by a former data science leader from the Israel Defense Forces. This acquisition aims to bolster trust and reliability in autonomous security systems. This acquisition is seen as a strategic move to enhance the testing, evaluation, and monitoring of machine learning systems and agentic workflows, according to Jonathan Zanger, Check Point’s Chief Technology Officer (CTO).

According to Zanger, the expertise that Deepchecks brings has grown increasingly important as generative artificial intelligence leads to new operational risks, such as the phenomenon known as “hallucinations,” where AI systems generate incorrect or nonsensical information. He emphasizes the pressing need for thorough evaluation mechanisms to ensure that AI models perform effectively and reliably.

During a recent interview, Zanger described the platform that Deepchecks has been developing for the past five years as particularly adept at assessing the performance of AI models. “They are data scientists who will understand how to evaluate models and build a system focused on testing the evaluation of monitoring solutions for agents,” he stated. This perspective underlines Check Point’s commitment to improving cybersecurity measures through AI technology.

Founded in 2019, Deepchecks is a relatively young firm that has quickly garnered attention. With a team of 15 and a successful seed funding round of $14 million led by Grove Ventures in June 2023, the company has been under the leadership of Philip Tannor, who has significant experience in the military’s research programs focused on advanced technologies.

Revolutionizing Network Security Management

Zanger highlighted that Check Point is actively re-evaluating what network security administration will mean in the coming years. The rise of autonomous AI systems necessitates new strategies for integrating these technologies into operational environments. The firm is exploring how AI agents will operate within enterprise networks and how security professionals will interact with them moving forward.

“The challenge is to ensure that AI agents are behaving correctly,” Zanger remarked. “We need to confirm that they operate within the right parameters, especially since their roles are often mission-critical.” His focus has been on finding a team with the necessary expertise to ensure that AI agents and models function as intended, thereby enhancing overall cybersecurity.

Check Point’s forthcoming platform aims to facilitate the management of complex network security landscapes. The design will incorporate AI-driven agents that support configuration, analysis, and operational tasks, while allowing human operators to maintain oversight. This innovative approach points to a future where AI agents play an increasingly collaborative role in cybersecurity tasks.

Zanger further elaborated on the platform’s potential: “When we integrate their knowledge system into our network security orchestration, we will be able to assure CISOs that the agents they delegate critical projects to, such as microsegmentation, will perform as expected.” This promise of reliability represents a significant advancement in how organizations can manage AI participation in security frameworks.

In sectors where precision is paramount, inaccuracies or unpredictable behavior from AI agents can have severe repercussions. Check Point is dedicated to mitigating these risks by ensuring that AI agents either deliver accurate results or clearly acknowledge when they do not possess the necessary information.

“When building models, it is crucial to have a reliable source of truth that can be monitored,” Zanger explained. “Agents should only either provide the correct answer or indicate their uncertainty.” This dual approach aims to enhance trust in the technologies implemented across security operations.

Validating AI Workflows and Outputs

Evaluating the effectiveness of enterprise or generative AI systems is complex, particularly when outputs can differ from one interaction to another while remaining contextually valid. Zanger noted that Deepchecks has developed unique intellectual property to assess these agentic workflows, ensuring that varying outputs can still lead to correct conclusions.

“In dealing with generative AI, outcomes can differ but still be accurate,” he clarified. “For instance, when querying ChatGPT, one might receive multiple responses that are all valid.” The challenge lies in consistently assessing the accuracy of these responses, a task that Deepchecks specializes in.

For AI agents to effectively protect critical assets within an organization, they must ask pertinent questions, access the right data sources, and continuously deliver reliable guidance. Zanger emphasized that Deepchecks’ technology will be pivotal in verifying these interactions, ensuring that AI systems behave predictably and accurately in crucial security contexts.

“We are keen on having Deepchecks devote their complete attention to what we consider a highly impactful challenge in cybersecurity,” Zanger stated, demonstrating enthusiasm for the collaboration and its potential to tackle pressing security issues.

Initially focused on enhancing network security orchestration, Deepchecks is poised to provide additional layers of validation and optimization across Check Point’s technological offerings. Zanger highlighted that applications related to threat prevention, threat intelligence, and AI capabilities will particularly benefit from this partnership, which aims to incorporate the validation technology into a broader portfolio.

Overall, the integration of Deepchecks is viewed as a crucial step toward establishing a robust framework for improving trust in AI technologies. Zanger believes this boost in confidence will encourage security practitioners to embrace new and disruptive technologies in the cybersecurity landscape.