Cequence Security reveals agent personas for enhanced AI agent management

Cequence Security has launched Agent Personas within its AI Gateway. The capability provides enterprises with infrastructure-level control over AI agent activity, aiming to address the privilege gap that identity verification alone may not fully resolve.

In many organisations, AI agents connect to enterprise applications using the Model Context Protocol (MCP). A common assumption is that identifying an agent’s identity is sufficient to control its actions. However, agents do not apply judgement when using available access. Agent Personas aims to address this by using plain-language job descriptions to define scoped virtual MCP endpoints for each agent role.

For example, a customer service AI agent is assigned CRM read-only access, while a coding agent can read GitHub issues and create Jira tickets but cannot merge pull requests. A CI/CD automation agent can access specific pipeline tools and a limited notification channel.

The release also introduces Agent Access Keys, a composite credential designed for headless agents operating in automated workflows. These keys combine agent identity, user identity, and persona-level privileges into a single traceable credential, aiming to provide forensic visibility for security teams.

Agent Personas features include:

• Scoped virtual MCP endpoint per agent role: Each persona defines access down to the API endpoint and permission level.
• Natural language persona creation: The gateway selects tools based on plain-language descriptions.
• Single source of truth: Updating a persona applies changes across all associated agents.
• Agent Access Keys: A composite credential for headless agent environments.
• Per-tool policy enforcement: Enables rate limits, data masking, and approval workflows at the tool-call level.
• Full audit trail: Every tool call is traceable to the agent, user, persona, and timestamp.
• Model-agnostic enforcement: Works across OpenAI, Google, Anthropic, open-source, and custom models.

The urgency is highlighted by figures indicating that more than 80% of Fortune 500 companies deploy AI agents, while 47% have AI-specific safeguards in place. Cequence’s capability is positioned to support organisations moving from pilot deployments to production-scale use.

Early deployments indicate use in complex enterprise environments. For example, a U.S. telecommunications company used Agent Personas to restrict agent access across tools such as GitLab, Confluence, Jira, and Slack, ensuring agents only accessed required resources and reducing lateral access risk.

Cequence states that this approach is intended to help organisations balance security, governance, and scalability as they deploy AI across customer, employee, and operational workflows.