Ahmed is an ARENA alumnus with experience across technical AI safety research, governance, and field-building. His work includes research surveying proposed solutions to DeepMind’s AI Safety Gridworlds and aspiration-based designs for safer intelligent agents.

 

He further strengthened his technical foundations through the Center for AI Safety’s Machine Learning Safety Scholars program and SPAR. Ahmed has supported 25+ interdisciplinary AI safety capstone projects for the Center for AI Safety and BlueDot Impact, ranging from formal control frameworks to international AI governance.

 

He is currently researching adversarial robustness for cyber defence at the University of Queensland and co-directs the Queensland AI Safety Initiative.

​

Kyle was the Learning Director at WhiteBox Research, where he led technical sessions and mentored fellows for two cohorts of the SEA-focused AI safety research fellowship. He also led program design for Condor Camp, Southeast Asia’s first AI safety camp in 2024.

 

Recently, he was a fellow at Pivotal Research during the Summer '25 cohort, investigating activation plateaus as an approach to find concept boundaries in LLMs. Before this, he worked on white-box probe subversion under the UK AISI bounty program and interpretability research for many-shot jailbreaking at SPAR in 2024.

Before AI safety, Kyle was a Founding Engineer at Haast, an AI for marketing compliance startup.

​

​

Ajay is a doctoral student pursuing research on building better evals for detecting strategic deception in frontier models. Particularly,  he is interested in how we can leverage insights from adversarial games using to detect regions in human brain responsible for social cognition for the purpose of deception eval building.

 

His previous research has involved sociolinguistics, computational social science, and recently mechanistic interpretability. 

 

Ajay has completed multiple AI safety courses such as ARENA, BlueDot and AISC, and collaborated with various international teams working to make superintelligence human-salient. He's looking forward to supporting everyone on their mission to leverage AI safety for the better.