Upwind Unveils AI Security Thesis

The big picture: Upwind today announced a new product direction and its “Security for AI” thesis, signaling a fundamental shift in how the company approaches AI risk management.

Why it matters:

• Attack Surface Shift: Traditional runtime security focused on process execution and network flows, but the critical threat activity has moved to the application layer, involving APIs, payloads, prompts, and thousands of multi-cloud platform (MCP) calls by AI agents.
• Invisible Threats: Issues like prompt injection, data leakage, and over-permissioned tool calls do not manifest in traditional packet monitoring, requiring a new approach to visibility.
• Inventory Crisis: The proliferation of AI consumption methods (managed services, self-hosted models, custom agents) creates a critical inventory problem, with security teams often lacking visibility into new AI components and their dependencies.

How it works:

• Integrated Security: Upwind’s thesis argues that AI security cannot be a standalone product but must be woven into every layer of cloud security, from code pipelines to runtime environments.
• AI Inventory Layer: The company’s updated platform provides an AI inventory layer that maps relationships, dependencies, and risks across AI components, including Bedrock Agents, Azure OpenAI Assistants, and self-hosted agents.
• Enhanced Scanning: Upwind is updating its scanning capabilities to address the increased velocity and complexity of AI-generated code, which introduces more code from more sources and faster dependency integration.

The catch: While Upwind’s integrated approach to AI security addresses critical gaps, the rapid evolution of AI models and agentic capabilities means security solutions must constantly adapt. The industry still largely treats AI security as a niche, potentially underestimating the pervasive nature of AI-related risks across the entire cloud stack. Competitors will likely follow suit, making comprehensive, integrated AI security a new battleground.