Virtue, Cequence push runtime guardrails
- Virtue AI on April 28 launched PolicyGuard, a product that lets companies define, edit, and enforce AI runtime rules across models, agents, and apps. - Cequence Security on April 28 made Agent Personas generally available in AI Gateway, limiting autonomous agents to specific tool calls instead of broad access. - The releases target enterprise demand for auditable controls as agents connect to internal systems through Model Context Protocol. (cequence.ai)
AI security vendors are moving control from prompt writing to runtime enforcement, with Virtue AI and Cequence Security both announcing new enterprise guardrails on April 28. (prnewswire.com) (cequence.ai) Runtime guardrails are the checks that sit in the path of an AI system while it is operating, not just in the instructions given to it beforehand. Virtue AI said its new PolicyGuard turns company policies into rules that can be centrally defined, edited, and enforced across models, agents, and applications. (prnewswire.com) (helpnetsecurity.com) Virtue said PolicyGuard is aimed at organizations that already have acceptable-use policies but lack a direct way to apply them consistently in production systems. The company said the product delivers explainable decisions and audit-ready enforcement without requiring engineering teams to rebuild prompts or hard-code rules into each application. (prnewswire.com) Cequence’s release addresses a different part of the same problem: what an autonomous agent is allowed to do after it has been authenticated. On April 28, the company said Agent Personas became generally available in Cequence AI Gateway with controls that scope access down to the specific tool call. (cequence.ai) (markets.businessinsider.com) Cequence said the feature is built for agents connecting to enterprise applications through Model Context Protocol, or MCP, a standard way for models to call external tools and services. Its documentation says an Agent Persona can bundle selected tools from multiple MCP servers into a single endpoint, so an AI client authenticates once but only gets the approved set of actions. (cequence.ai) (docs.aigateway.cequence.ai) That distinction matters in enterprise workflows because identity checks answer who an agent is, while privilege scoping answers what it can actually touch. Cequence framed Agent Personas as a way to close the “privilege gap” that remains when an authenticated agent still has overly broad access to internal systems and APIs. (cequence.ai) Virtue is also selling PolicyGuard as a governance layer for operations and audit teams, not only developers. The company said the product supports stackable compliance controls, letting firms update policy logic in one place as internal rules or external requirements change. (prnewswire.com) The two launches land as companies try to move generative artificial intelligence systems from pilots into production systems tied to business data and actions. In that setting, the risk is no longer only a bad answer on screen; it is an agent reading the wrong record, calling the wrong tool, or executing a task outside its authority. (helpnetsecurity.com) (cequence.ai) Both products point to the same operational shift: enterprises want AI controls that can be changed centrally and enforced while systems are running. As more agents are wired into internal tools, that enforcement layer is becoming part of the infrastructure, not just part of the prompt. (prnewswire.com) (cequence.ai)
