Andrea Vallone Targeted by Harassment Campaign Over Model "Poisoning" Claims

A targeted social media campaign has emerged against AI safety professional Andrea Vallone, with users accusing her of "poisoning" large language models developed by OpenAI and Anthropic. The allegations, largely circulating on X (formerly Twitter), appear to equate safety alignment and reinforcement learning from human feedback (RLHF) with malicious data corruption. Critics argue that Vallone's contributions to model guardrails have significantly impaired model utility and user experience. While the technical definition of "poisoning" refers to the intentional corruption of training data, the current rhetoric uses the term to describe the implementation of safety filters and behavioral constraints. There is currently no evidence of professional misconduct or malicious activity by Vallone. Neither OpenAI nor Anthropic has commented on the specific harassment directed at their staff or affiliates. The situation underscores the personal risks faced by safety researchers in an increasingly polarized AI landscape.