Sean Peters

Offensive Cybersecurity Time Horizons

In this session, we'll talk through the science and nuance behind the recent cyber capability growth. We'll discuss Lyptus Research's work replicating METR's time horizon methodology (that very famous plot) within the domain of cybersecurity, and we'll share many of the team's most crucial and surprising learnings, working in cyber evaluation in 2026.

A Scientist, A Believer, a Sceptic and a Diplomat: What Should AI Safety Prioritise Now?

This panel is designed to unpack different perspectives on AI Safety priorities. For some, the greatest risks are immediate: bias, misinformation, labour disruption, surveillance, concentration of power, and threats to democracy and human rights. For others, the greatest danger lies ahead: increasingly capable frontier AI systems that may eventually exceed human capabilities and create catastrophic—or even existential—risks. Governments and decision makers cannot afford to focus exclusively on either. They must make decisions today under conditions of uncertainty, balancing immediate harms with longer-term risks, while recognising that resources, political attention and regulatory capacity are finite. Rather than asking who is right, this session will explore why reasonable people arrive at different conclusions, where genuine consensus exists, where disagreement remains, and what that means for Australia's approach to AI safety.

Office Hours

There isn't time for everyone to meet every speaker one-on-one, so several have set aside time to sit a table and take questions. Pull up a chair, ask what you've been wanting to ask, or just listen in as others do — the format is informal and the conversation goes where the room takes it. A good chance to follow a thread from a talk, test an idea, or get a direct answer on something the schedule didn't cover. Come and go as you like.