Contact Us

EY and Microsoft Put AI Agent Accountability at the Center of Cyber Defence

Gaming Is Becoming a More Demanding Technology Business

EY and Microsoft are extending their alliance into a part of cybersecurity where speed alone is no longer the main question. EY has launched its Agentic Cyber Defence Centre, an initiative that combines Microsoft security technologies with EY’s cyber expertise, policy-based controls, and tamper-evident records of AI agent activity.

The ambition is to connect defences that often operate across separate enterprise systems while keeping people responsible for decisions made with automation. Rather than presenting AI agents as a replacement for security teams, EY describes an operating model in which agents can assist with work under defined authority.

That distinction matters. As organizations give AI systems access to tools and operational processes, they must determine not only what an agent can detect, but also what it is permitted to do.

EY Global Consulting Cybersecurity Leader Richard Watson argues that the next stage of cybersecurity calls for a different operating model, not merely a faster security operations center. EY’s approach combines its cyber intellectual property, EY.ai Value Blueprints, and agentic accelerators with Microsoft’s security platform, threat intelligence, and agentic technology.

Cybersecurity professionals monitoring enterprise security systems and automated threat detection dashboards in a modern security operations center.

Are you looking for engineers?

Cybersecurity professionals monitoring enterprise security systems and automated threat detection dashboards in a modern security operations center.

According to EY, the centre applies policy controls to agent actions and maintains tamper-evident records. These records are intended to help teams review activity and understand what changed during a security response.

A record that shows an action was taken is not proof that the action was appropriate. Its usefulness depends on whether teams can examine the evidence, understand the policy behind the decision, and intervene when necessary.

Consider an agent that identifies suspicious account activity. Detecting an anomaly is different from disabling an account or changing access permissions. Those steps may affect legitimate employees and critical business functions. A sound operating model distinguishes actions that can proceed automatically from those requiring approval or escalation.

The EY announcement outlines an intended approach to this problem. It does not establish that every proposed control has produced a measured reduction in risk or operating costs.

The centre is designed around security domains that are frequently connected in practice but managed through different technologies: identity, data, applications, cloud, and operational technology.

An incident involving one compromised credential, for example, may also involve an application, a data repository, and a cloud service. Effective coordination requires information to move between systems without losing context or weakening access restrictions.

Are you looking for engineers?

For companies adopting AI agents, governance cannot sit only in a policy document. It needs to be reflected in backend permissions, secure API interactions, reliable data pipelines, auditability, monitoring, and software testing.

Engineering teams must also plan for change. When an application, model, or access rule is updated, automated workflows should be tested again to ensure that they still respect the intended boundaries. Human reviewers need enough information to challenge a recommendation rather than merely approve it.

This work often demands capacity beyond a company’s existing development team. Square Codex, a Costa Rica-based nearshore software development and staff augmentation partner serving North American companies, can support custom applications, backend systems, API integration, cloud development, data engineering, and QA automation.

Those services can help organizations build and maintain the software surrounding AI-enabled workflows without suggesting that Square Codex provides EY’s cyber defence platform or guarantees security outcomes.

EY and Microsoft’s announcement points to a shift in the question enterprises must ask about agentic cybersecurity. Speed and analytical capability matter, but so do permission boundaries, traceable decisions, and the ability to stop or review an action. In systems that can act, accountability has to be engineered into the process.

Cybersecurity professionals monitoring enterprise security systems and automated threat detection dashboards in a modern security operations center.

Are you looking for engineers?

Cybersecurity professionals monitoring enterprise security systems and automated threat detection dashboards in a modern security operations center.

For companies adopting AI agents, governance cannot sit only in a policy document. It needs to be reflected in backend permissions, secure API interactions, reliable data pipelines, auditability, monitoring, and software testing.

Engineering teams must also plan for change. When an application, model, or access rule is updated, automated workflows should be tested again to ensure that they still respect the intended boundaries. Human reviewers need enough information to challenge a recommendation rather than merely approve it.

This work often demands capacity beyond a company’s existing development team. Square Codex, a Costa Rica-based nearshore software development and staff augmentation partner serving North American companies, can support custom applications, backend systems, API integration, cloud development, data engineering, and QA automation.

Those services can help organizations build and maintain the software surrounding AI-enabled workflows without suggesting that Square Codex provides EY’s cyber defence platform or guarantees security outcomes.

EY and Microsoft’s announcement points to a shift in the question enterprises must ask about agentic cybersecurity. Speed and analytical capability matter, but so do permission boundaries, traceable decisions, and the ability to stop or review an action. In systems that can act, accountability has to be engineered into the process.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top