Cequence Security has launched Agent Personas in its Cequence AI Gateway, providing enterprises with detailed infrastructural control over AI agent operations. This development offers granular control, addressing the critical privilege gap which cannot be resolved through identity verification alone, enhancing security measures for organisations deploying AI agents through the Model Context Protocol (MCP).
As AI agents increasingly connect to enterprise applications, it is crucial to understand that simple authentication does not equate to monitoring agent actions. Unlike human users, AI agents lack the discretion to limit their access privileges. Agent Personas manages these risks by employing plain-English descriptions to define a scoped virtual MCP endpoint for each specific agent role.
Virtual MCP endpoint
The introduction of Agent Personas allows customisation of access permissions for various AI agents. For instance, a customer service AI may access CRM data in a read-only capacity, while a coding agent can interact with GitHub issues and Jira tickets without modifying pull requests. Similarly, CI/CD automation agents are restricted to particular pipeline tools and a single notification channel.
The release also features Agent Access Keys, a new type of composite credential specifically designed for headless agents in automated workflows. These keys consolidate agent identity, user identity, and persona-level privileges, providing security teams with enhanced forensic abilities to determine agent actions and capabilities at any time.
Agent personas capabilities
- Scoped virtual MCP endpoint per agent role, defining access down to specific API endpoints presented as a single endpoint.
- Natural language persona creation, enabling plain language descriptions for agent tasks while the gateway selects appropriate tools.
- Single source of truth, allowing persona updates to apply instantly across all agents using it without code modifications.
- Composite Agent Access Keys binding agent identity, user identity, and persona privileges for headless agents.
- Per-tool policy enforcement with rate limits, data masking, and approval workflows at individual tool call levels.
- Comprehensive audit trail showing each tool call's attribution to specific agents, users, personas, and timestamps.
- Model-agnostic enforcement across various platforms, including OpenAI, Google, Anthropic, open-source, and custom models.
Complexity of securing agents
Securing AI agents is particularly crucial, as highlighted by Gartner®, noting that the focus should go beyond model runtime and inference security.
It should include the complexity of managing agent actions, suggesting that agents should demonstrate their authority before gaining access to tools and data. Statistics show that while over 80% of Fortune 500 companies utilise AI agents, only 47% have AI-specific security measures in place.
Investments in AI platforms
Early adopters of Agent Personas exhibit the capability to regulate AI agent access within complex systems
Early adopters of Agent Personas exhibit the capability to regulate AI agent access within complex systems. A prominent U.S. telecommunications firm used this technology to manage agent interactions across platforms like GitLab, Confluence, Jira, and Slack, ensuring agents have limited access only to necessary resources, eliminating risks of lateral access.
Ameya Talwalkar, CEO and Co-Founder at Cequence, stated, "Enterprises have made massive investments in AI, and the race to put agents into production across customer experiences, employee workflows, and business operations is accelerating fast. However, security, governance, and scale requirements can’t be ignored. Cequence closes the gap that has been holding organisations back by automatically limiting agent tool access which lowers costs, enhances performance, and improves security."
Powering customer commerce
Shreyans Mehta, CTO and Co-Founder at Cequence, remarked, "AI agents have quickly become a channel as significant as the web or mobile, powering customer commerce, employee productivity, and autonomous operations all at once. Agent Personas is how you govern infrastructure access at the agent level, enforcing exactly what each agent can do down to the specific API endpoint, across any model or platform. It is the control plane enterprises need to confidently and securely enable agentic AI access to applications and data."
Agent Personas expands upon the existing Cequence AI Gateway, which has surpassed 140 verified enterprise application integrations, protects over 10 billion daily API interactions, and safeguards four billion user accounts. Cequence's influence is further recognised through its ranking on the Deloitte Technology Fast 500 and its leadership in the KuppingerCole Leadership Compass for API Security and Management.
Cequence Security, the pioneer in application, API, and Agentic AI protection, announces the general availability of Agent Personas in Cequence AI Gateway. Agent Personas give enterprises granular, infrastructure-level control over what AI agents are permitted to do — down to the specific tool call — closing the critical privilege gap that identity alone cannot address.
As organisations deploy AI agents to connect to enterprise applications via the Model Context Protocol (MCP), a dangerous assumption has taken hold: that authenticating who an agent is amounts to controlling what it can do. It does not.
Virtual MCP endpoint
Agents inherit the privileges of their users and, unlike humans, have no judgement about when not to use available access. Agent Personas solves this by using a plain-English job description to define a scoped virtual MCP endpoint for each agent role.
For example, a customer service AI agent gets CRM read-only access, not the ability to modify records. A coding agent can read GitHub issues and create Jira tickets but cannot merge pull requests. A CI/CD automation agent gets access to specific pipeline tools and a single notification channel, and nothing else.
Single attributable credential
The release also introduces Agent Access Keys, a new composite credential type purpose-built for headless agents running in automated workflows. Each key binds agent identity, user identity, and persona-level privileges into a single attributable credential, giving security teams the forensic clarity to answer exactly who did what, when, and under which permissions.
Agent Personas capabilities include:
- Scoped virtual MCP endpoint per agent role: each persona defines access down to the specific API endpoint and permission level, presented as a single endpoint
- Natural language persona creation: describe in plain language what an agent should do; the gateway selects the right tools automatically
- Single source of truth: update a persona once and the change applies immediately across every agent using it, with no code changes required
- Agent Access Keys: a composite credential binding agent identity, user identity, and persona privileges into a single attributable credential for headless agents
- Per-tool policy enforcement: rate limits, data masking, and approval workflows applied at the individual tool call level
- Full audit trail: every tool call is attributable to the specific agent, user, persona, and timestamp
- Model-agnostic: enforced at the infrastructure layer across OpenAI, Google, Anthropic, open-source, and custom models equally
Complexity of securing agents
The urgency is significant. According to Gartner®, “While the current market focus is on model runtime and inference security, this approach fails to address the complexity of securing agents, where the primary risk is not what the AI says, but what the AI does.” It further states, “If an AI agent cannot prove who it is acting for and why, it should not get access to tools and data.”
More than 80% of Fortune 500 companies now deploy active AI agents, yet only 47% have AI-specific safeguards in place. Enterprises have made substantial investments in AI platforms but have struggled to move agents from pilot to production safely. Agent Personas is built to change that.
Substantial investments in AI platforms
Early deployments show how Agent Personas govern AI agent access in complex environments. One major U.S. telecommunications provider used the technology to prevent agents from crossing boundaries in tools like GitLab, Confluence, Jira, and Slack. Scoped virtual endpoints ensured each agent accessed only what it needed, eliminating lateral access risks without extra infrastructure.
“Enterprises have made massive investments in AI, and the race to put agents into production across customer experiences, employee workflows, and business operations is accelerating fast,” said Ameya Talwalkar, CEO and Co-Founder at Cequence. “However, security, governance, and scale requirements can’t be ignored. Cequence closes the gap that has been holding organisations back by automatically limiting agent tool access which lowers costs, enhances performance, and improves security.”
Powering customer commerce
“AI agents have quickly become a channel as significant as the web or mobile, powering customer commerce, employee productivity, and autonomous operations all at once,” said Shreyans Mehta, CTO and Co-Founder at Cequence.
“Agent Personas is how you govern infrastructure access at the agent level, enforcing exactly what each agent can do down to the specific API endpoint, across any model or platform. It is the control plane enterprises need to confidently and securely enable agentic AI access to applications and data.”
Agent Personas builds on strong momentum for Cequence AI Gateway, which has grown to more than 140 verified enterprise application integrations. Cequence protects more than 10 billion daily API interactions and four billion user accounts across its platform — recognised as No. 128 on the 2025 Deloitte Technology Fast 500, a Leader in the 2025 KuppingerCole Leadership Compass for API Security and Management, co-chair of TM Forum's AI-Native Blueprint Initiative on Agentic Interaction Security, and co-author of three CIS Critical Security Controls® Companion Guides for securing AI agents, LLMs, and MCP environments alongside the Centre for Internet Security and Astrix Security.
