Sydney · 2026
I build tools to reduce real world AI harms
I'm Dan, a software engineer at Canva working on AI safety and privacy. Outside work I focus on technology-facilitated abuse and AI safety: non-consensual intimate imagery, family-violence contexts, image provenance, and evals that make safeguards testable.
Now focused · safety research engineering Now studying · technology-facilitated abuse and AI safety Now cooking · rosemary focaccia Now focused · safety research engineering Now studying · technology-facilitated abuse and AI safety Now cooking · rosemary focaccia
AI can cause psychological and social harm to people and those around them. I build systems that make abuse harder and decisions safer.
That means practical tools for provenance and abuse prevention, as well as evals that make safeguards easier to inspect and improve.
see the work Research experiments
Adversarial safety evals
- Best-of-N StrongREJECT eval An LLM refusal-robustness experiment measuring ASR-vs-N under a fixed StrongREJECT subset, query budget, and model judge. LLM refusal · Evals
- Text-only T2I jailbreak eval A text-to-image safety-filter experiment separating filter bypass from moderation-confirmed harmful success. T2I filters · Evals
Lately
Out of the kitchen
Have a look around