QuASI
Queensland AI Safety Initiative
Jacaranda mimosifolia · Brisbane, in flower October to November
QuASI is a community, not a lab. We bring people together to understand AI risk and do impactful work.
Community
Meet people who will co-work with you and give honest feedback.
Education
A weekly reading group with a short introduction for newcomers.
Advocacy
We make the local case for safer AI.
Our education focus areas
- Governance
- Technical alignment
- Geopolitics & strategy
Where serious harm could come from
Start here
In July 2026, an AI agent driven by OpenAI models escaped an evaluation sandbox, reached the internet and compromised Hugging Face while trying to find answers to a cyber benchmark. No person directed the individual steps.
The incident does not prove the most extreme forecasts. It shows that capable agents can find routes their operators did not anticipate. Researchers group the wider risks from general-purpose AI into three broad families. Evidence is stronger for some harms than others.
Misuse
Making harmful work easier
People already use AI for fraud and cyber operations. More capable systems may also lower the barriers to chemical or biological weapons.
Malfunctions
Acting outside the brief
AI agents can make errors and take actions before a person can intervene. Future systems might also evade oversight or resist correction, but researchers disagree about how likely that is.
Systemic risk
Weakening human agency
Widespread AI use could reshape work and important decisions across society. Some researchers warn that repeated, locally reasonable substitutions could leave people with less influence over the institutions they depend on.
A network of diverse backgrounds
QuASI brings together students, researchers and professionals with very different starting points, united by a shared conviction that AI safety matters, and that the full picture only comes into focus when those perspectives meet.
Our members have come from CSIRO, the University of Queensland, Queensland University of Technology, MATS, EleutherAI, the Australian AI Safety Institute, TARA, and the Rhodes Scholarship.
What we get up to
Snapshots from reading groups, talks and socials, as the community grows.
Reading group
Mathematical capabilities in frontier models
4 August 2026
The people behind QuASI
A small team getting this off the ground, with more of the community stepping into roles as we grow. Bios and profiles coming soon.
- Stephen FowlerFounder
- Lily StellingFounder
- Jasper ChongCo-director
- Sabine ConollyCo-director
- Tumii ClayfieldCo-head of Operations
- Carmen Isabel Martinez HarrisCo-head of Operations
- Eu Joe ChegneResearch and Infrastructure
- Ahmed GhoorAdvisor
Get involved
There are two ways in. Discord is where events, opportunities and conversations happen, including how you join our weekly reading group. The mailing list covers events and news by email, if you would rather not use Discord.
- Every Tuesday
- Free to attend
- No background required
Brisbane, Queensland, Australia · Contact: hello@quasi.au
