Uploaded August 2026 | Updated September 2026, 2 weeks ago
In this interview from Black Hat 2026, Joe Hladik, head of Zero Labs at Rubrik, joins theCUBE's Krista Case to discuss a newly discovered AI vulnerability that allowed researchers to break out of Microsoft Copilot's sandbox and reach backend Azure infrastructure. Hladik explains why Rubrik Zero Labs prioritizes backup data as an intelligence source, treating it as an untapped signal alongside traditional endpoint and network telemetry. He details how the team discovered the flaw in February, followed responsible disclosure with Microsoft and confirmed a patch by mid-March, while noting the underlying exploit technique could resurface in other AI copilots. Hladik also touches on similar vulnerabilities found in AI orchestration platforms, including Langflow and ChatMate, underscoring how quickly AI tooling is expanding the enterprise attack surface.
The conversation also explores why AI excels at the "science" of a problem but struggles with the tradecraft and creativity attackers bring, making the combination of skilled humans and powerful models the real danger. Hladik breaks down the observability challenge facing defenders, arguing that logs alone aren't enough — security teams must build their own baselines of normal AI agent behavior to spot abnormal activity. He shares a CISO's framework for prioritizing vulnerabilities based on existing security layers rather than treating every finding as critical, and revisits fundamentals like maintaining and actively using a software bill of materials. Hladik provides a practical roadmap for security leaders navigating an AI threat landscape that is evolving faster than detection tools can keep pace.
Find more SiliconANGLE news and analysis siliconangle.com
Follow theCUBE's wall-to-wall event coverage siliconangle.com/events
Learn about the latest theCUBE events thecube.net
00:00 - Intro
00:07 - Unveiling Cyber Threats: AI Vulnerabilities and Attack Innovations
04:31 - Novelty and Impact of the AI Vulnerability Discovery
06:58 - AI and Human Collaboration: Strengths and Limitations
09:02 - Expanding Research Beyond Microsoft: AI Orchestration Risks
11:49 - Navigating AI Evolution: The Role of Context and Expertise in Cybersecurity
15:06 - Strategic Risk Management and Future Directions in Security
#theCUBE #BlackHat #theCUBEresearch #Rubrik #AI #CyberSecurity #CloudSecurity
In this interview from Black Hat 2026, Joe Hladik, head of Zero Labs at Rubrik, joins theCUBE's Krista Case to discuss a newly discovered AI vulnerability that allowed researchers to break out of Microsoft Copilot's sandbox and reach backend Azure infrastructure. Hladik explains why Rubrik Zero Labs prioritizes backup data as an intelligence source, treating it as an untapped signal alongside traditional endpoint and network telemetry. He details how the team discovered the flaw in February, followed responsible disclosure with Microsoft and confirmed a patch by mid-March, while noting the underlying exploit technique could resurface in other AI copilots. Hladik also touches on similar vulnerabilities found in AI orchestration platforms, including Langflow and ChatMate, underscoring how quickly AI tooling is expanding the enterprise attack surface.
The conversation also explores why AI excels at the "science" of a problem but struggles with the tradecraft and creativity attackers bring, making the combination of skilled humans and powerful models the real danger. Hladik breaks down the observability challenge facing defenders, arguing that logs alone aren't enough — security teams must build their own baselines of normal AI agent behavior to spot abnormal activity. He shares a CISO's framework for prioritizing vulnerabilities based on existing security layers rather than treating every finding as critical, and revisits fundamentals like maintaining and actively using a software bill of materials. Hladik provides a practical roadmap for security leaders navigating an AI threat landscape that is evolving faster than detection tools can keep pace.
Find more SiliconANGLE news and analysis siliconangle.com
Follow theCUBE's wall-to-wall event coverage siliconangle.com/events
Learn about the latest theCUBE events thecube.net
00:00 - Intro
00:07 - Unveiling Cyber Threats: AI Vulnerabilities and Attack Innovations
04:31 - Novelty and Impact of the AI Vulnerability Discovery
06:58 - AI and Human Collaboration: Strengths and Limitations
09:02 - Expanding Research Beyond Microsoft: AI Orchestration Risks
11:49 - Navigating AI Evolution: The Role of Context and Expertise in Cybersecurity
15:06 - Strategic Risk Management and Future Directions in Security
#theCUBE #BlackHat #theCUBEresearch #Rubrik #AI #CyberSecurity #CloudSecurity
