Home

Professor Ufuk Topcu presented at The Exchange: AI + Autonomy, hosted by the Special Competitive Studies Project (SCSP) on September 25, 2025, in Austin, TX. The event convened leaders from government, industry, and academia to discuss how artificial intelligence and autonomy are shaping national security.

Mustafa Karabag and Prof. Ufuk Topcu from The Center for Autonomy at UT Austin found that large language models excel at inferring hidden information but struggle to withhold it, often revealing too much in social deduction games like The Chameleon. Their findings expose a key weakness in AI’s strategic reasoning, particularly in adversarial or high-stakes scenarios where discretion is critical.

A new study co-authored by researchers from The Center for Autonomy introduces the “LLM Brain Rot Hypothesis,” showing that large language models can lose reasoning ability when repeatedly trained on low-quality, engagement-driven web content such as short social-media posts. The work demonstrates measurable declines in ethical behavior, reasoning, and long-context performance, signaling that model quality can erode even without changes to architecture or scale.