MPCH is partnering with UK-based cybersecurity provider Socura to test ATLAS, its AI Authorization Control Plane, in security operations where AI agents can take actions such as isolating devices, blocking user accounts, or changing system settings. The collaboration targets a growing enterprise security concern: how to prevent autonomous AI systems from executing high-impact actions without explicit human authorization. Rather than focusing only on detecting malicious activity, ATLAS is designed to enforce cryptographic authorization before an AI-initiated action reaches execution.
ATLAS Adds an Authorization Layer Between AI Decisions and Execution
ATLAS is designed to sit between an AI agent’s proposed action and the system that carries it out. Operators define the actions an agent is permitted to perform, while the control plane checks whether a requested operation falls within those boundaries. The goal is to separate an agent’s ability to recommend or initiate an action from the authority to execute it.
That distinction is particularly relevant in managed detection and response (MDR). Security operations centers increasingly use automation to investigate alerts and contain threats. Actions such as quarantining an endpoint, disabling an account, or changing a security configuration can limit damage during an incident, but an incorrect response can also interrupt business operations or lock legitimate users out of critical systems.
MPCH says ATLAS applies cryptographic authorization before execution, creates signed records of actions, and can stop an action mid-execution through a kill switch if it exceeds defined boundaries. It also supports cold-signed approvals, with authorization keys held offline in air-gapped secure storage. Under this model, an AI agent can request a sensitive action but cannot independently generate the authorization needed to approve it.
The architecture is intended to keep high-consequence decisions under human authority while allowing lower-risk operations to proceed within predefined policies. MPCH’s product description also says ATLAS can intercept requests, enforce policy, validate outputs, and generate signed records for execution events.
Socura provides managed detection and response services through a UK-based security operations center. Its work includes monitoring networks, endpoints, and cloud environments, investigating alerts, threat hunting, and containing incidents. The company supports technologies including endpoint detection and response, network detection and response, SIEM, and security orchestration, automation, and response platforms.
That makes its operations a relevant environment for testing an authorization control plane. Security teams must balance response speed with the risk of acting on incomplete or misleading information. An AI agent may correctly identify suspicious activity but still make a poor containment decision if it lacks business context or misinterprets an alert.
The MPCH-Socura collaboration is intended to evaluate whether ATLAS can enforce approval requirements around those actions. However, the announcement describes a testing collaboration, not a completed deployment or independently validated security result. It does not disclose the test scope, duration, technical integration details, or performance measurements.
The Broader Challenge: Controlling Agentic Privileges
The partnership reflects a wider shift in enterprise cybersecurity. Traditional security controls focus on protecting infrastructure, managing identities, detecting malicious activity, and restricting access. Agentic AI introduces another layer of risk: software systems that can interpret information, select tools, and execute actions across connected environments.
An agent with access to security tools may have permissions to investigate alerts and perform containment operations. If those permissions are too broad, a compromised agent, manipulated instruction, or incorrect decision could trigger actions beyond the operator’s intentions.
This is where authorization becomes distinct from authentication and monitoring. Authentication establishes which identity or system is making a request. Authorization determines what that identity is allowed to do. Monitoring records activity and helps security teams investigate it. ATLAS is designed to add a cryptographically enforced authorization checkpoint to the execution process.
MPCH’s approach also highlights the importance of separating approval credentials from the AI runtime. Cold-signed approvals are intended to keep sensitive signing authority outside the agent’s environment, reducing the possibility that an agent or remote attacker could independently approve restricted operations. Signed execution records could also help organizations reconstruct what happened during an incident.
The design does not eliminate the need for other security controls. Organizations would still need to secure agent identities, manage credentials, validate tool integrations, protect the authorization infrastructure, and review the policies governing permitted actions. A human approval mechanism can also introduce operational delays if approval thresholds are too restrictive or escalation procedures are poorly designed.
From Cryptographic Infrastructure to AI Governance
MPCH is bringing its experience in cryptographic infrastructure into the emerging market for agent security through ATLAS and its defense subsidiary, Ti22. The company says Ti22 has supported U.S. Department of War operations and government institutions internationally. Its broader product portfolio includes cryptographic key management and secure data-transit technologies alongside AI authorization.
For Socura, the collaboration provides an opportunity to examine how authorization controls fit into an operational MDR environment, where automation must work alongside human analysts and existing security platforms. The company has also been expanding its AI-enabled security operations capabilities through its work with Palo Alto Networks.
The commercial significance will depend on whether ATLAS can integrate with different agent frameworks and security tools without adding excessive latency or complexity. It will also need to demonstrate that authorization policies remain enforceable during interruptions, system failures, and attempts to manipulate an agent’s instructions.
The MPCH-Socura test is therefore an early step toward addressing a specific weakness in agentic AI deployments: the gap between an AI system’s ability to act and the authority it should have to do so. If the approach proves effective in operational testing, cryptographic authorization could become an important component of enterprise controls for autonomous security systems and other high-impact AI applications.
Market Landscape
The emerging market for AI agent security is expanding beyond model safety and conventional access management. Organizations deploying agents that can take consequential actions need controls across several layers:
- Agent identity: Establishing which agent or workload is requesting an operation.
- Authorization: Restricting actions according to policies, risk levels, and human approvals.
- Execution security: Preventing unauthorized commands from reaching connected systems.
- Observability and auditability: Maintaining verifiable records of decisions and execution events.
- Human oversight: Escalating sensitive operations to authorized people rather than allowing agents to approve their own actions.
ATLAS focuses on the authorization and execution-control layers. Its proposed cryptographic approvals and signed records distinguish it from tools that primarily monitor agent behavior or detect suspicious activity.
Socura provides the operational environment for testing these controls, particularly in MDR workflows where automated containment can have immediate consequences for customers.
The broader competitive challenge is interoperability. Agent security controls must work across different models, frameworks, tools, identity systems, and enterprise environments. They must also preserve response speed without weakening authorization requirements.
Top Insights
- MPCH and Socura are testing ATLAS in security operations to prevent AI agents from executing sensitive actions without human authorization.
- ATLAS introduces cryptographic policy enforcement between an agent’s requested action and its execution.
- Cold-signed approvals are designed to keep sensitive authorization keys outside AI systems and remote execution environments.
- A cryptographic kill switch is intended to stop actions that exceed operator-defined boundaries.
- The partnership remains a testing initiative; no independent performance results or completed production deployment have been disclosed.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
