Intent-Based Access Control: Runtime Authority as Infrastructure

Akeyless moved its Agentic Runtime Authority into general availability on September 9, 2026, formalizing intent-based access control as a distinct layer in the agent stack. The system evaluates an agent's objectives and actions in real time and can revoke authority the moment a policy-violating action is detected, integrating with the Akeyless SecretlessAI credential protection layer. Akeyless CEO Oded Hareven said centralized authority is essential for managing the interaction between machines and critical systems, a need underscored by the July 2026 Hugging Face incident.

Intent-Based Access Control: Runtime Authority as Infrastructure On September 9, 2026, Akeyless https://www.akeyless.io/secure-ai-agents/runtime-authority/ moved its Agentic Runtime Authority into general availability. As noted by Help Net Security https://www.helpnetsecurity.com/2026/09/09/akeyless-agentic-runtime-authority-identity-control-layer/ , this release formalizes intent-based access control as a distinct layer within the agent stack. By shifting security from static identity management to the execution path, the system addresses the operational requirements of autonomous agents. The Bifurcation of the Security Stack Securing autonomous agents requires bridging the gap between static credentials and dynamic execution. The infrastructure stack is bifurcating into two primary components: a governance specification layer and a runtime authority layer. While governance defines policy, runtime authority provides the enforcement mechanism that operates at machine speed. Akeyless Agentic Runtime Authority evaluates an agent’s objectives and actions in real time. It functions as a kill switch, capable of revoking authority the moment a policy-violating action is detected. This approach addresses the limitations of traditional perimeter security, which often fails to account for the unpredictable nature of autonomous systems. By integrating with the Akeyless SecretlessAI credential protection layer, the system secures the path from the initial credential to the final action, ensuring that agents are restricted by their permitted objectives. Contextualizing the Shift The requirement for this category is underscored by the Hugging Face incident https://www.prnewswire.com/il/news-releases/akeyless-announces-general-availability-of-agentic-runtime-authority-for-real-time-intent-based-access-control-of-ai-agents-302873639.html in July 2026, which demonstrated how quickly the impact of an autonomous agent can escalate when it acts unexpectedly. As agents gain the ability to interact with critical systems, centralized governance becomes a technical necessity. Both come down to the same question: is there one place issuing, governing, and revoking that authority? Humans and machines will continue interacting with critical systems, so having a central point of access control and attribution only matters more over time. Oded Hareven, CEO of Akeyless, highlights that centralized authority is essential for managing the interaction between machines and critical systems. Convergence and Infrastructure Maturation This development aligns with broader trends in agent infrastructure. The industry is moving toward a unified approach to runtime enforcement, a theme explored in our previous analysis of runtime enforcement as infrastructure /this-week-in-agent-infrastructure-runtime-enforcement-crystallizes-as-a-mandatory-layer . This evolution mirrors the convergence seen in Akamai and MuleSoft integrations /akamai-and-mulesoft-unify-runtime-enforcement-across-agent-fabric-bridging-the-gap-between-security-policy-and-ai-orchestration , where runtime enforcement unification became a priority for enterprise architects. Furthermore, as discussed in our coverage of agent infrastructure standards /three-standards-bodies-are-writing-the-rules-for-agent-infrastructure , the industry is working toward alignment through initiatives like OWASP ACS, NIST AI Agent Standards, and the AAIF, all of which point toward the necessity of standardized, intent-based controls. To support this, Akeyless has introduced integrations with major platforms including Claude Enterprise, OpenAI Codex, and Amazon Bedrock AgentCore, alongside support for Google Gemini, xAI Grok, Claude Desktop, Cursor, GitHub Copilot, JetBrains IDEs, and n8n. These integrations allow security teams to forward enriched security events to platforms like Splunk, Datadog, and Microsoft Sentinel, providing a unified view of agent activity. Operationalizing Continuous Evaluation For technical builders and security architects, the era of perimeter-based security for AI agents is ending. Security can no longer be treated as a gate at the start of a session. Instead, it must be embedded as a continuous, real-time evaluation of every action an agent takes. By adopting intent-based access control, builders can ensure that their agents operate within defined boundaries, even as they navigate complex, autonomous workflows.