Evo is built to respond to a transformed threat landscape powered by the rise of non-deterministic, autonomous systems. Traditional security models, which often rely on static rules or anticipatory scanning, are deemed inadequate by Snyk because they cannot keep pace with the dynamic behaviour of agentic systems. Snyk cites Gartner figures claiming one in three security leaders have experienced prompt-injection attacks and that 52 percent of organisations acknowledge unsanctioned internal AI tool deployment.
The platform introduces a layered agent-based architecture that uses autonomous Task Agents alongside a central Workflow Agent to coordinate operations from a single natural-language prompt. Task agents include Discovery Agent, Red-Teaming Agent, AI Risk Registry Agent and Policy Agent.
Snyk said the Workflow Agent orchestrates both Snyk-native and non-Snyk agents, bridging previously disjointed tooling environments. The platform is built around the OODA loop—originally adapted from military strategy—to enable continuous and adaptive protection rather than periodic assessments.
Key features include an AI Bill of Materials scan to inventory all AI components within a codebase, automated red-teaming of agentic flows, runtime scanning of model-serving compute platforms, natural-language policy creation, and integrated reporting tools. These capabilities signal how Snyk anticipates enterprises will need to govern the full stack of AI-native software—from model development and tool integration through to production runtime.
Despite the ambition, Evo is currently offered in an experimental preview for early design-partners, with wider availability scheduled for early 2026. Enterprises eager to adopt the platform must apply via the Snyk website.
Security analysts have noted that while the demand for AI-native application security is rising, many organisations still lack mature governance structures for agentic systems. For example, academic research on architectures such as SAGA emphasises the need for lifecycle-level controls, including identity, delegation and runtime oversight, which align with Snyk’s approach.
The platform release comes as organisations face a surge of AI-related risks — including prompt injection, data leakage, model exfiltration and tool misuse — compounded by increased complexity of software supply-chains. Snyk argues that conventional AppSec tools were created for deterministic applications and cannot adapt quickly enough for systems that learn, delegate and act autonomously.
However, critical questions remain. Organisations will need to evaluate whether the agentic orchestration itself introduces new attack surfaces. The market for AI-native security is still nascent, and integration with existing legacy systems, developer workflows and compliance frameworks may present significant challenges. Some developers may also question whether outsourcing orchestration to a central agent controller reduces transparency or complicates incident response.
Topics
Technology