Blueprint Alliance: Unified Standards for Securing AI Agents
Amidst the rapid expansion in the adoption of AI Agents within work environments, Docker has revealed its participation as a founding member of the Blueprint Alliance. This alliance aims to address security gaps and agent governance through a reference framework that brings together 12 leading technology companies, including AWS, CrowdStrike, Okta, and Google Cloud.
Why does this alliance matter?
Enterprises today face a dual challenge: either allowing AI adoption without sufficient security controls or hindering innovation. Historically, the industry has proven that technical collaboration is the optimal solution for overcoming adoption complexities, as seen with PCI DSS standards in the financial sector or CIS in cybersecurity. The alliance aims to move AI agents from the phase of "every team building its own solutions" to the phase of unified standards.
Alliance goals and security pillars
The alliance provides a framework based on four fundamental governance questions: Where are the agents located? What are their permissions? What are they actually doing? And how can one respond when an emergency occurs?
In parallel, the Agent Baseline initiative was launched, defining 35 open and documented security controls covering six core outputs: discovery, restriction, authorization, monitoring, verification, and response. These controls aim to bridge the gap between potential threats and the security capabilities required in enterprises.
Docker's role in securing the runtime environment
While other alliance members focus on external governance (identity and access), Docker focuses on "internal" governance, i.e., the agent's execution environment. The company asserts that relying on protection via APIs alone provides a false sense of security. Therefore, Docker provides solutions such as Docker Sandboxes to ensure agent isolation, and Docker Hardened Images to secure the software supply chain, ensuring that security policies are enforced at the runtime environment level itself.
These standards allow enterprises to evaluate their existing infrastructure against unified benchmarks, accelerating the pace of secure and sustainable AI adoption.