Max McWhae
I work on technical AI safety from Perth, Western Australia — and on making sure Perth has an AI safety community at all.
I'm building SANDGLASS, a study of evaluation awareness and sandbagging in open-weight language models: whether models internally represent “I am being evaluated,” and whether deliberate underperformance triggered by that recognition can be caught. I completed BlueDot Impact's Technical AI Safety course, I'm working through the ARENA curriculum, and I'm studying mathematics at UWA. I've been writing code since 2002. Everything I build ships open — including the negative results.
Research
SANDGLASS — measuring evaluation awareness and detecting trained sandbagging on consumer hardware. In progress.
Community
Founder and organiser of the Perth AI Safety Meetup — relaunching as a monthly series at UWA in semester 2, 2026. Group page link coming soon.
Elsewhere
Writing
- 2026-07-28 This site is live