Jarrah Bloomfield

About

I work on security at Anthropic, previously at Google. I care about using information security to have a positive impact on the world.

Writing

• Information Security in High-Impact Areas — 80,000 Hours career review
• Retbleed Explainer — an accessible breakdown of the Retbleed vulnerability
• Celiacs Don't Need to Live in Fear — a risk-based approach to managing celiac disease
• Donutting is Bad — why the "unlocked computer" prank has hidden costs

Acknowledgements

• Toy Models of Superposition — Anthropic, 2022
• Red Teaming Language Models to Reduce Harms — Anthropic, 2022
• Language Models (Mostly) Know What They Know — Anthropic, 2022
• Constitutional AI: Harmlessness from AI Feedback — Anthropic, 2022
• The Capacity for Moral Self-Correction in Large Language Models — Anthropic, 2023
• Frontier Safety Framework — Google DeepMind, 2024

Games

• Too Many Cooks — a party deduction game for 2-6 players
• Know Your Pants — a puzzle game

Links