https://gkanellopoulos.com/Recent content on George KanellopoulosHugoen-usCopyright © 2026, George Kanellopoulos.Wed, 15 Jul 2026 07:00:00 +0200https://gkanellopoulos.com/ai-in-the-enterprise/ai-governance-needs-architecture-not-better-tools/Wed, 15 Jul 2026 07:00:00 +0200https://gkanellopoulos.com/ai-in-the-enterprise/ai-governance-needs-architecture-not-better-tools/The AI governance tool market is booming. The governance gap is getting worse. Those two facts are connected. Organizations are buying platforms to solve a problem that requires architecture.https://gkanellopoulos.com/ai-in-the-open/step-change-world-isnt-ready-for/Wed, 08 Jul 2026 10:00:00 +0200https://gkanellopoulos.com/ai-in-the-open/step-change-world-isnt-ready-for/The UN convened its first Global Dialogue on AI Governance in Geneva and its new Scientific Panel delivered a straightforward preliminary report: agentic AI is a governance step change, and the window to act is closing. The panel’s evidence is clear, but whether multilateral institutions can build governance architecture that matches the systems it must govern remains the open question.https://gkanellopoulos.com/ai-in-the-enterprise/the-effective-agent/Thu, 02 Jul 2026 10:00:00 +0200https://gkanellopoulos.com/ai-in-the-enterprise/the-effective-agent/Agentic AI is the most hyped and least operationalized technology of 2026: 79% of organizations report adopting AI agents, yet only 11% have solutions in production. This white paper examines why, through the anatomy of a production harness, the architecture patterns that survive contact with reality, the ways agents fail differently than traditional software, and the governance, economics and organizational readiness that determine who closes the gap. It ends with ten recommendations for technical leadership.https://gkanellopoulos.com/ai-in-the-enterprise/the-operation-was-successful-but-the-patient-died/Thu, 25 Jun 2026 09:22:03 +0200https://gkanellopoulos.com/ai-in-the-enterprise/the-operation-was-successful-but-the-patient-died/The term hallucination creates a dangerous false impression in enterprise settings. A simple expert-driven framework for evaluating LLM outputs on two axes: completeness and accuracy.https://gkanellopoulos.com/enterprise-architecture/are-you-ai-i-am-ea/Thu, 18 Jun 2026 10:00:00 +0200https://gkanellopoulos.com/enterprise-architecture/are-you-ai-i-am-ea/Why probabilistic systems demand architectural governance, not better procedures, and why that puts Enterprise Architecture back at the center.https://gkanellopoulos.com/ai-in-the-open/tool-metadata-poisoning-ctms/Thu, 02 Apr 2026 10:00:00 +0200https://gkanellopoulos.com/ai-in-the-open/tool-metadata-poisoning-ctms/Nothing verifies that a tool’s description is what its publisher wrote, and researchers have repeatedly demonstrated poisoning attacks against MCP clients. CTMS is a signing and verification scheme that prevents these attacks at runtime.