Dr. Charalambos Theodorou

Founder & CEO of Xybern | PhD in Artificial Intelligence | Agentic Security

I am the Founder and CEO of Xybern, an agentic cybersecurity company building the authorisation and security infrastructure required for autonomous AI agents to operate safely across enterprise and mission-critical environments.

My work sits at the intersection of artificial intelligence, autonomous agents, cybersecurity, identity and authorisation with a focus on building the trust infrastructure required for AI agents to safely access systems, data, APIs, tools and other agents.

Explore Xybern About Me Research & Publications

Building the Authorisation Infrastructure for AI Agents

Artificial intelligence is moving from systems that generate information to autonomous agents capable of making decisions and taking actions.

These agents can call APIs, access databases, interact with enterprise applications, connect to MCP servers, invoke tools, trigger transactions and delegate work to other agents.

As their autonomy increases, one fundamental security question becomes increasingly important:

The Decision

What should an AI agent be allowed to do and under what conditions?

Through Xybern, I am working on the authorisation infrastructure required to answer that question before an autonomous action reaches execution.

Every agent action should have explicit, verifiable authority behind it.

Every agent action is intercepted before it executes and evaluated against your charter. Xybern returns one of three verdicts, allow, escalate or block, in a single synchronous call, then seals the outcome to the Provenance Vault.

finance-copilot · payments.transfer

Deterministic rulesallow-list

Semantic intentmismatch

Risk verdict62 / 100

Escalated to human rcpt_9f2a…c71

How it decides

The action passes through the evaluation stack in order: deterministic rules first, then semantic intent, then a signed risk verdict. The first decisive check short-circuits, so obvious cases resolve in microseconds and only the ambiguous ones reach a model.

Where it runs

The same decision is enforced wherever the action originates, inside your agent through the SDK, at the MCP and AI gateway, or over a plain HTTP call. One charter, one verdict, every path.

Fail-closed by default

If the layer cannot reach a verdict, the action does not run. A high-stakes call never falls through to trust because a check was momentarily unavailable.

Explore Xybern

Latest Milestone

Xybern & Blossom Accelerator

Xybern has been selected to join the ecosystem of Blossom Accelerator in Saudi Arabia, marking an important milestone in the company's growth and its expansion within the region. The milestone supports Xybern's mission to build the trust, security and authorisation infrastructure required for the emerging agentic economy and strengthens the company's presence within Saudi Arabia's rapidly developing artificial intelligence ecosystem.

Read the Announcement → Visit Xybern →

Founder & CEO of Xybern

Autonomous AI agents can already interact with production systems, access sensitive data, call external tools, initiate transactions and delegate authority to other agents.

Traditional identity and access-management systems can authenticate who or what is making a request, but autonomous agents introduce a more difficult problem.

Explore Xybern

The Platform

Xybern is the authorisation layer between AI intent and execution.

Should this specific agent be permitted to perform this specific action, using these resources, under these conditions, right now?

Xybern is designed to make that decision before execution.

The platform sits between autonomous agents and the systems they interact with, intercepting actions and evaluating identity, policy, permissions, delegation, context and risk before determining whether an action should be authorised or denied.

Runtime Authorisation

Every protected agent action is evaluated before it executes.

Xybern checks the requested action against active policies, agent authority, delegation context and environmental conditions before issuing an authorise or deny decision, creating an enforceable security boundary between an agent's intent and its impact.

Agent Identity

Every registered autonomous agent can receive a distinct identity so actions can be attributed to the specific agent or workload that initiated them.

Identity becomes the foundation for determining authority, permissions and accountability throughout the system.

Agent-to-Agent Authorisation

Modern AI systems increasingly involve multiple autonomous agents delegating tasks and authority to one another. Xybern controls how that authority can move between agents.

Delegations can be explicitly scoped, time-limited and revoked, while child agents are prevented from escalating beyond the authority of the originating agent. The full delegation chain remains auditable.

MCP & Tool Security

MCP gives AI agents access to tools and external capabilities. Xybern provides the authorisation boundary around that access. Organisations can control:

which agents may connect to a particular MCP server;
which tools an agent may invoke;
which actions are permitted;
which resources may be accessed;
which parameters or transaction values are acceptable;
how long authority remains valid;
whether authority can be delegated;
and which actions require human approval.

Every protected tool call can be evaluated before it runs.

Fine-Grained Policy Enforcement

Xybern allows organisations to define precisely what an autonomous agent may and may not do.

Policies can govern individual actions, resources, tools, parameters, transaction limits, workflows and delegation boundaries. Rather than relying only on broad credentials or static role assignments, authorisation can be evaluated at the point of execution.

Human Approval for High-Risk Actions

Not every autonomous action should execute without human involvement.

High-risk actions can be paused and escalated to an authorised human before execution, allowing organisations to maintain autonomy for routine operations while introducing explicit approval boundaries around sensitive or consequential actions.

Cryptographic Provenance

Authorisation should be provable, not simply logged.

Xybern records authorisation decisions, delegation chains, approvals and execution outcomes as verifiable security evidence. Its provenance architecture is designed to provide a tamper-evident record of why an action was permitted or prevented and which authority governed the decision.

External Agent Federation

Enterprises increasingly need to interact with agents developed by partners, vendors and external organisations.

Xybern allows external agents to enter an environment without automatically inheriting trust. The organisation maintains control over which external agents are recognised, what resources they can access, which tools they may invoke, whether authority can be delegated further and which actions require approval.

Flexible Deployment

Xybern is designed to work across different enterprise and regulatory environments. Deployment options can support cloud, dedicated and sovereign architectures depending on security, regulatory, isolation and data-residency requirements. Integration can be achieved through different patterns:

API gateways;
SDK integration;
MCP gateways;
agent-to-agent enforcement;
and policy-enforcement points.

Xybern works with existing AI infrastructure rather than requiring organisations to replace their models, agents or frameworks.

Framework & Model Agnostic

The authorisation layer is designed to work independently of the underlying model or agent framework.

Organisations can govern autonomous systems built with different models, orchestration frameworks and enterprise AI stacks through the same authorisation infrastructure.

Explore Xybern

Why Agentic Security Matters

The security model for autonomous AI is fundamentally different from the security model for traditional software. Software normally executes predefined logic.

Autonomous agents can reason about objectives, choose tools, decide which actions to perform, create or delegate tasks and interact dynamically with external systems. That introduces new security challenges.

Agentic Security

Why Agentic Security Matters

An authenticated agent is not necessarily an authorised agent.
A valid credential does not necessarily mean every action performed with that credential should be permitted.
A delegated task should not automatically transfer unlimited authority.
Access to an MCP server should not imply unrestricted access to every tool exposed by that server.
And a log showing that an action occurred does not necessarily prove why that action was authorised.

This is the security problem I am focused on through Xybern:

Creating enforceable authority boundaries for autonomous AI.

Explore Xybern

My Work

My work spans artificial intelligence research, autonomous systems and the security infrastructure required to deploy intelligent systems safely in real-world environments.

Agentic Security

My current work focuses heavily on the emerging security requirements of autonomous AI agents, including agent identity, runtime authorisation, delegation, MCP security, policy enforcement, provenance and human oversight.

Artificial Intelligence

My broader research and engineering background covers artificial intelligence, machine learning, deep learning, large language models, reinforcement learning and intelligent systems.

Autonomous Agents

I am particularly interested in how autonomous agents reason, interact with tools, coordinate with other agents, delegate authority and operate safely within enterprise environments.

Applied AI Research

A core part of my work involves translating artificial intelligence research into practical architectures, products and systems capable of operating in real-world environments.

PhD in Artificial Intelligence

I hold a PhD in Artificial Intelligence and Machine Learning.

My academic and professional work has covered areas including machine learning, computer vision, deep learning, reinforcement learning, large language models, autonomous systems and artificial intelligence security.

My research background continues to shape how I approach the technical and security challenges created by increasingly autonomous AI systems.

Explore my research & publications

Research & Publications

Research has been an important part of my career and continues to influence how I approach technology, entrepreneurship and product development. My interests span Artificial Intelligence, Machine Learning, Computer Vision, Deep Learning, Autonomous Systems, Large Language Models, AI Agents and Agentic Security.

I have contributed to academic publications, patents and applied artificial intelligence research across multiple areas of AI and autonomous systems. Today, a major area of my focus is the intersection between autonomous artificial intelligence and cybersecurity particularly the infrastructure required to establish identity, authority, policy and accountability for AI agents.

View research & publications

Insights

I write about the transition from generative AI to autonomous agents and the security infrastructure that this transition will require AI agent authorisation, agent identity, agent-to-agent delegation, MCP security, autonomous systems, agentic cybersecurity and the broader agentic economy.

Read all insights

September 2026 · News & Updates

Xybern, Blossom and Saudi Arabia: The Next Chapter

Xybern joining Blossom Capital's inaugural portfolio represents more than a company milestone. It is an opportunity to build the security infrastructure for autonomous AI in one of the world's most ambitious emerging technology ecosystems.

About Dr. Charalambos Theodorou

I am an AI researcher, engineer and entrepreneur and the Founder & CEO of Xybern. My background combines academic artificial intelligence research, commercial AI engineering and entrepreneurship.

I hold a PhD in Artificial Intelligence and have worked across machine learning, computer vision, large language models, autonomous systems and AI security. Today, my primary focus is Xybern and the security infrastructure required for autonomous AI agents to operate safely across enterprise, regulated and mission-critical environments.

I am particularly interested in the emerging intersection between artificial intelligence and cybersecurity as AI evolves from systems that generate information into autonomous actors capable of executing actions in the real world.

Read my full biography

Media & Speaking

I speak, write and contribute to discussions around the technologies and security challenges shaping autonomous artificial intelligence topics including AI Agents, Agentic Cybersecurity, AI Agent Authorisation, Agent-to-Agent Security, MCP Security, Artificial Intelligence, Autonomous Systems, AI Identity & Trust and The Agentic Economy.

For interviews, media enquiries, speaking opportunities, research discussions and industry collaboration:

Get in touch

Stay Connected

Follow my work on artificial intelligence, autonomous agents, agentic security and Xybern.

LinkedIn Google Scholar Xybern