Tag
The paper introduces Aegis, a runtime governance system for agentic AI that mediates tool actions through trusted authorization, preventing risky side effects in evaluated sandbox scenarios.
Agentao introduces a governed local-first runtime for tool-using LLM agents, separating model-generated actions from host-authorized execution to improve safety and governance.
This paper presents Canonical Action Verification and Attestation (CAVA), a runtime-semantics layer that converts heterogeneous agent activity into canonical runtime action objects for governance, enabling consistent approval binding, receipt reproducibility, and policy enforcement across diverse AI execution environments.
This paper introduces Base Sequence Analysis, a framework that encodes LLM agent runtime behavior into compact sequences, revealing high-risk patterns like the 'P-X-P' trigram and a verification deficit. It presents Governor, a runtime intervention system that improves task success by 6.2% and reduces token consumption by 44%.
Microsoft released the Agent Governance Toolkit, an open-source runtime enforcement tool for AI agents that provides deterministic policy enforcement, zero-trust identity, and sandboxing, covering all 10 OWASP Agentic risks with over 13,000 tests.
The article argues that companies are overinvested in AI intelligence (model capability) while neglecting crucial runtime layers for authority, accountability, and reality representation, leading to potential failures when AI acts within institutions.
The article discusses the need for runtime governance in AI agents to balance autonomy with compliance, introducing SAFi, an open-source framework that enforces policies in real-time and audits actions.
The article argues that prompt-based alignment methods face an architectural ceiling, proposing a 'Runtime Governance Layer' with hard constraints between generation and execution, drawing parallels to biological self-preservation and Terror Management Theory.