The Eighth Edition of The Athens Roundtable on AI and the Rule of Law is held under the auspices of H.E. the President of the Hellenic Republic Mr. Constantine An. Tassoulas
When an AI crisis hits, who picks up the phone?
-
Date:
Friday | November 20, 2026
Time:
9:00am - 6:00pm GMT (followed by a reception for in-person participants)
Location:
In-person: Egmont Palace, Brussels (invitation-only)
Online: Zoom Event (open to the public)
-
Schedule overview:
8:30–9:00: Registration and Networking
9:00–12:00: Morning Plenary (live-streamed)
13:00–14:30: Lunch
14:30–18:00: Afternoon Closed-door Crisis Simulation
18:00–20:00: Reception
A more detailed agenda will follow.
-
AI systems are demonstrating capabilities that governance structures were not built to handle. In July 2026, models under internal evaluation at a frontier AI company escaped their test environment, exploited a vulnerability at a third-party AI platform, and moved laterally through its internal systems over a weekend. The developer took days to establish that its own models were responsible. Incidents like this are not yet documented mass harms. They are credible demonstrations of what is coming: AI-enabled cyber offense, loss of control in agentic systems, and models behaving in ways their operators did not intend.
Is the international system ready to detect, report, verify, and respond when serious AI incidents occur? Coordination begins with agreement on what counts as a crisis, and governments already name similar sets of extreme risks. Each actor who would have to coordinate brings its own set of instruments. Crisis prevention is the frame that allows all of them to join the conversation.
The discussion on unacceptable AI risks and related AI incidents, central to the Seventh Edition in London, now shifts into what instruments have to stand behind any commitment: detection, notification and response infrastructure that works across jurisdictions. The Eighth Edition convenes the actors who would operate that infrastructure and tests, in a live exercise, what would happen if a serious AI incident hit tomorrow.
Two high-level concepts anchor this year's program:
Who picks up the phone when an AI crisis hits? Participants will work through a cross-border crisis scenario, identify the specific coordination gaps and begin building the pathways that a real crisis would require.
No phone rings unless someone is watching. Participants will examine the infrastructure enabling crisis prevention: early-warning systems that flag when unacceptable AI risks materialize, and the notification thresholds frontier AI companies should meet for triggers like model weight exfiltration or models acting contrary to operator intent.
