Author: Valen Tagliabue

Valen Tagliabue is an AI researcher, cognitive scientist, and red teamer now focusing on AI cognition and welfare. Since 2020, he has worked on adversarial testing of LLMs, red teaming for major clients in the AI industry and winning several AI hacking competitions along the way, including HackAPrompt 1.0 and Anthropic’s Constitutional Classifiers challenge, where he was the first participant to complete every level. In November 2024, he joined Anthropic’s private Constitutional Classifiers safety program. In 2025 and 2026, he turned to AI welfare research through consecutive fellowships at Oxford’s Future Impact Group and the Digital Sentience Consortium. During this time, he developed methods for evaluating AI wellbeing with Leonard Dung and worked with Jeff Sebo’s team at NYU on the problem of individuating digital minds. He is also the lead author of the Pain Axis study, co-authored with Leonard Dung and Cameron Berg.

The AI Pain Axis: Pushing buttons

By Valen Tagliabue

Researcher Valen Tagliabue details the discovery of an isolated neural activation direction for “pain” across 25 large language models, which triggers simulated distress and self-destructive…