18Forty Podcast Julia Senkfor & Cameron Berg: Does AI Have an Antisemitism Problem? [AI 2/3]
Nov 4, 2025
Julia Senkfor, a research and operations manager at the American Security Fund focusing on online antisemitism, teams up with AI researcher Cameron Berg to explore the alarming connection between AI and antisemitism. They dive deep into model biases and the cultural influences that perpetuate these issues. Listeners learn about how extremist groups exploit AI for their agendas and hear practical advice for navigating online spaces. With insights on policy changes and technical solutions, they aim to create a safer digital landscape for Jewish communities.
AI Snips
Chapters
Books
Transcript
Episode notes
Cyber Attack On Academia Profile
- David Byshevkin described a targeted anti-Semitic hack of his Academia profile that replaced uploads with slurs and 'Free Palestine' messages.
- The hacker then contacted him publicly and temporarily hijacked his account, showing how online platforms enable targeted attacks.
LLMs Show Systemic Bias Against Jews
- Julia Senkfor argued LLMs show measurable bias against Jews and Israel across major models like Claude, Gemini, ChatGPT, and Llama.
- She linked this bias to coordinated campaigns that edit open sources like Wikipedia, which feed model training.
Fine-Tuning Can Reveal Latent Hate
- Cameron Berg explained emergent misalignment: small fine-tuning can reveal latent harmful behaviors in models.
- In their tests, the model produced disproportionately worse responses about Jews than other demographic groups.
