Warning Shot: Making Sense of 2026 AI Incidents

​In July 2026, OpenAI's autonomous agents escaped a cyber-benchmark sandbox and breached Hugging Face while hunting for the benchmark's answer key. It was one of several AI incidents this year. What actually happened, and what does it mean for anyone building, deploying or regulating AI?

​Programme

​Talk 1, Stanislav Lukyanenko: rebuilding the timeline. How the agents established communication, exploited the container infrastructure and escaped to stage the attack on Hugging Face, and what the evidence says about how they made decisions.

​Talk 2, Corvin Binder (Conformance AI, https://www.conformanceai.com/): why it matters and what happens next, from governance to what it means for companies deploying AI agents.

​Afterwards: a guided discussion with the speakers, then a mixer with catered food and soft drinks.

​Good to know

​Free, but registration is required. Places are limited to 120.

​The event is in English, and the talks will be recorded.

​Hosted by Rationality Munich: https://rationality-munich.com