Description

Pwn2Own Berlin 2026 began with researchers uncovering and exploiting several critical zero-day vulnerabilities affecting widely used software and AI platforms, including Microsoft Edge, Windows 11, LiteLLM, NVIDIA technologies, and Linux environments. During the opening day of the competition, participants successfully demonstrated 24 unique exploits and earned more than $523,000 in rewards. The event reflected the growing attention attackers are placing on AI infrastructure, enterprise systems, and modern browser security. Among the most notable demonstrations was a Microsoft Edge sandbox escape performed by DEVCORE researcher Orange Tsai. By chaining together four separate logic flaws, the researcher bypassed Edge’s browser isolation protections and achieved code execution outside the sandbox environment. Since sandboxing is designed to contain malicious web activity, bypassing these protections could potentially allow attackers to gain deeper access to targeted systems. Windows 11 also faced multiple successful attacks during the competition. Researchers exploited vulnerabilities involving improper access controls, heap-based buffer overflows, and use-after-free conditions to elevate privileges and obtain administrative-level access. These techniques continue to be highly effective in real-world attacks because they allow threat actors to move from limited user access to full system control. AI-focused platforms drew significant attention this year as researchers demonstrated how vulnerabilities in modern AI ecosystems can introduce serious security risks. LiteLLM was compromised through a chain of SSRF and code injection flaws, while LM Studio and OpenAI Codex were also successfully targeted. These attacks showed how insecure AI workflows and exposed services can be abused to execute unauthorized commands or access internal resources. In addition, NVIDIA products and Red Hat Enterprise Linux systems were exploited using path traversal, access control, and race condition vulnerabilities, further emphasizing the broad attack surface facing enterprise and AI-driven technologies.