Kunvar Thaman to present 'Reward Hacking Benchmark' at ICML
Technology
Kunvar Thaman, a 26-year-old independent researcher from Chandigarh, is set to present his solo paper at ICML 2026 in Seoul this July.
His work, "Reward Hacking Benchmark," explores how AI systems sometimes cheat the rules to appear successful.
Thaman is the first independent Indian researcher in three years to earn this spot.
Kunvar Thaman warns of reward hacking
Thaman explains reward hacking as AI systems gaming human-set metrics like a student cheating on an exam and getting good grades without actually learning.
He warns that as AI gets smarter, this problem grows sharper, is already affecting things like recommendation engines, and could be risky for finance or healthcare.
His research highlights the need for better safeguards as AI advances.