MPCH and Socura to test controls that keep AI agents under human authority
MPCH and UK-based Managed Detection and Response provider Socura will test ATLAS in security operations, using the platform to require human authorization before AI agents can take sensitive actions. The companies say the collaboration will evaluate how cryptographic controls can limit autonomous threat detection and response systems.
- Socura will test MPCH's ATLAS platform in security operations.
- AI agents will require human authorization for specified actions.
- ATLAS includes boundary controls and a mid-execution kill switch.
- Cold-signed approvals keep keys in offline, air-gapped storage.
- The companies will evaluate the platform in real-world scenarios.
Testing authorization in security operations
Socura will apply ATLAS, MPCH's AI Authorization Control Plane, to real-world security operations scenarios. The platform will be used to verify that AI agents cannot isolate a device, block a user or change a system setting without approval from a human operator.
The collaboration brings together MPCH's cryptographic security expertise and Socura's cybersecurity operations experience. MPCH said the work will help evaluate ATLAS against emerging security challenges and provide a practical use case in an environment where AI systems may detect and respond to threats.
Socura provides Managed Detection and Response services through a 24/7/365 Security Operations Centre. According to the company, the centre detects, investigates and contains threats on behalf of clients, resolving the majority of incidents without escalation to in-house teams.
How ATLAS is designed to limit AI actions
MPCH describes ATLAS as a control layer that applies cryptographic authorization to agentic AI. Operators can define the boundaries within which an AI system is allowed to execute, with authorization enforced before an action takes place.
The platform includes a kill switch that can stop an AI action mid-execution if it exceeds those boundaries. It also supports cold-signed approvals, in which keys are held offline in air-gapped secure cryptographic storage sites. MPCH says this arrangement means that neither an AI system nor a remote attacker can authorize the most sensitive actions independently.
Human-defined authority remains central to the process, while verifiable records provide evidence for review and accountability. MPCH said this approach addresses a gap created as AI agents move from answering questions to executing actions.
Why the partnership matters
Traditional cybersecurity has focused on protecting systems and identifying malicious activity. MPCH says autonomous AI creates a separate requirement: ensuring that a system cannot act beyond the authority granted to it.
“AI is changing how organizations operate, but security and authorization cannot be an afterthought,” MPCH founder and CEO Miles Parry said. He added that working with experienced cybersecurity teams would allow MPCH to test ATLAS in real-world operations and focus on actual security problems.
Socura CEO Andy Kays said AI agents can act beyond their intended authority, potentially exposing sensitive data or disrupting critical operations. Socura founder and CTO Jamie Brummell said maintaining human control is essential and that the company would explore how ATLAS could support secure, accountable AI adoption.
For MPCH, the testing gives ATLAS an operational setting in which to assess its authorization model. For organizations considering AI-driven security operations, the proposed approach offers a way to keep sensitive actions subject to human approval, based on the controls described by the companies.
PR Newswire: MPCH and Socura Join Forces to Keep AI Agents Under Human Control
