Safety & Security

Nine CISO-led groups target autonomous security operations

Nine CISO-led groups under Edward G Amoroso will build guidance for AI security operations centers, covering models, telemetry, governance, human duties and vendor interoperability.

Agentic SOC Alliance launches nine CISO-led groups
Agentic SOC Alliance launches nine CISO-led groupsjarmoluk / pixabay
By Marcus Bennett5 min read

Why it matters

  • The Agentic SOC Alliance launched nine CISO-led working groups.
  • Dr Edward G Amoroso will coordinate the effort after 31 years at AT&T.
  • The announcement names seven areas, including SOC Reference Architecture, Human Layer and Validation.
  • ExtraHop founded the alliance earlier this year; CrowdStrike, LangChain and ExtraHop are cited members.
  • The proposed architecture combines context, a harness control layer and interchangeable AI models.

By Marcus Bennett

The Agentic SOC Alliance launched nine CISO-led (chief information security officer) working groups to develop guidance for autonomous security operations centers (SOCs), with Dr Edward G Amoroso coordinating the coalition's effort.

The alliance's announcement names seven of those groups: SOC Reference Architecture, Human Layer, Agentic AI Layer, Infrastructure, Data and Context, Validation, and Industry Engagement. The announcement does not identify the other two.

Dr Edward G Amoroso serves as Chief Executive Officer of TAG Infosphere and as an NYU research professor. The alliance says he will oversee the work after 31 years at AT&T, where he served as Senior Vice President and Chief Information Security Officer.

The alliance brings together enterprise security leaders, researchers, academics, systems integrators and technology providers to define how an artificial intelligence (AI)-driven SOC should function. Its announcement cites CrowdStrike, LangChain and ExtraHop among the members.

ExtraHop founded the alliance earlier this year to define and standardize a common architecture for AI-driven security operations. The alliance says interest quickly grew among security leaders seeking to test the model and involve their own suppliers and teams.

What architecture is the alliance building?

The alliance describes a new SOC operating architecture built around context, a control layer called the harness, and interchangeable AI models. Its stated aim is to give autonomous agents real-time evidence while constraining actions through governance, guardrails, meaning limits on agent behavior, and human approval.

Agentic AI refers here to software agents that perform detection, response and analysis tasks. The alliance identifies three unresolved deployment questions:

  • The alliance says organizations must decide how much authority to grant autonomous agents.
  • The alliance says enterprises need methods to verify actions taken by those agents.
  • The alliance says organizations must ensure agents work safely across multiple tools and data sources.

“The shift to autonomous security operations is happening now, and enterprises need an architecture they can trust as agents begin taking actions with significant consequences,” said Dr Edward G Amoroso, Chief Executive Officer of TAG Infosphere and coordinator of the Agentic SOC Alliance.

Amoroso said the winning architecture must remain open and governed as agents, models and security technologies evolve. He said the alliance could make that architecture practical by building it, validating outcomes alongside industry veteran security leaders and creating a path to trusted production-scale autonomous security operations.

The alliance intends its reference architecture, meaning a common blueprint for system components, to guide production deployment and investment. The alliance says the working groups cover model behavior, human duties, evidence, workflows and testing as enterprises move beyond pilot projects.

Why does the alliance need a shared framework?

Several alliance members said no single supplier can answer the broader questions raised by AI-led security operations. The operating model must connect security products, agent behavior, governance and evidence flows across enterprise deployments.

“AI is finally letting defenders solve problems we've never been able to solve in the history of cyber defense, but trusted autonomy is the foundation we as an industry need to build together first,” said Greg Clark, Chief Executive Officer of ExtraHop.

Clark said the alliance offered an open reference architecture as a starting point. Within days, security leaders wanted more detail and brought vendors and teams together to create operational prototypes, he said. With Amoroso coordinating and CISOs shaping direction, Clark said the industry could build on an open, reliable and interoperable architecture for years.

Jim Routh, Chief Trust Officer at Saviynt, said agents require access to models, context and telemetry, plus tooling, orchestration, security controls, governance and trust. He argued that these interconnected layers must interoperate, meaning work together, reason together and operate safely. Without a common approach, he said, autonomy can become a liability.

Which SOC functions will the working groups define?\n

The alliance says the working groups cover several layers of SOC design and operation. Its seven named groups divide those layers as follows:

  • The alliance assigns the SOC Reference Architecture group the overall framework and publication of documents and tools.
  • The alliance assigns the Human Layer group tasks that remain with analysts: detection engineering, case ownership, threat hunting, meaning proactive searches for intrusions, and SOC management.
  • The alliance assigns the Agentic AI Layer group the mechanics of AI agents supporting or carrying out detection, response and analysis.
  • The alliance assigns the Infrastructure group orchestration, meaning coordination of agent workflows, and workflow management.
  • The alliance assigns the Data and Context group management of telemetry, meaning machine-generated security data, and intelligence feeds.
  • The alliance assigns the Validation group testing through red teaming, or simulated adversary testing, penetration testing and vulnerability management.
  • The alliance also lists an Industry Engagement group, completing the seven areas named in the announcement.

That structure matters for production because the alliance's model must connect machine action, analyst responsibility and test evidence. Its architecture treats models as interchangeable components rather than the operating architecture itself.

How will existing security systems fit?\n

Mark Thornberry, Chief Business Officer of GuidePoint Security, said the working groups will help security leaders bring AI into the SOC alongside functions they already run. He said the groups will address integration of SIEM (security information and event management), EDR (endpoint detection and response), NDR (network detection and response) and IAM (identity and access management).

They will also address how enterprises obtain AI models and applications, and how they measure results against industry peers, Thornberry said. The groups' output will form one Agentic SOC framework, he said. Leaders can then plan upgrades and investments against a shared standard.

What could make the framework usable across deployments?\n

The alliance says its output could help vendors and integrators align deployments across different customer environments. Chris Konrad, Vice President of Global Cyber at World Wide Technology, said Agentic SOC will reach production across enterprises when vendors and integrators build from one framework.

Konrad said the company is working with the working groups to design an overall reference architecture that holds across multiple customer environments. He said the coalition will validate that design against the company's experience of how enterprises actually deploy security operations.

The alliance's next milestone is one Agentic SOC framework that combines its reference architecture, validation work and guidance for human and industry layers. Konrad said production-scale adoption depends on vendors and integrators building from that common design, while Thornberry said enterprises need it to plan upgrades and investments against a shared standard.

Source: itbrief.com.au

Share this article:

More from Marcus Bennett

Marcus Bennett

Show full bio

Senior reporter covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

745 articles

Related articles

  1. Coalition plugs OpenAI, Anthropic, Gemini and Cursor into Wirespeed
  2. Google says 2027 is the year AI agents reshape enterprise security
  3. IDC names SAS a Major Player in agentic AI platforms
  4. IBM and Microsoft Launch Imagine Enterprise for Industry AI Agents
  5. Onyx and Glean integrate controls for enterprise AI agents

« Previous articleNext article »