What happened
OECD published the working paper 'Agentic AI in organisations: Early insights from practitioner interviews' on 16 September 2026 (AI papers series). It synthesizes semi-structured interviews with 25 organizations across 11 countries (developers, enterprise users, government agencies, research bodies) on agentic AI development/deployment choices, benefits, risks, and governance mechanisms, with none reporting unrestricted autonomy and evaluation/cybersecurity/accountability found underdeveloped.
Why it matters
This is authoritative multi-country primary evidence on how agentic AI is actually governed and where practitioner gaps (evaluation, security, human oversight) sit — it informs the OECD AI Principles and national policy, giving security teams a benchmark against which to compare their own agent controls.
Action needed
Read the paper to map its identified gaps (evaluation, cybersecurity, accountability, human oversight) against existing agent-security controls; use it as input to AI governance programs.